Seagate Announces First Hybrid Hard Drive

writertype writes "Today, Seagate announced about a dozen new products, including its first hybrid laptop hard drive that includes a 256-Mbyte flash chip to save power and speed up the time a notebook recovers from hibernation. Interestingly, the new Momentus 5400 PSD has also exceeded earlier estimates of hybrid hard-drive performance, which said that such drives would add an extra hour to the typical battery life of a notebook PC."

Comments

Tax deduction?

[Alex P Keaton in da]

Will these qualify me for a tax deduction?

Re:Tax deduction?

[Iguru42]

Only if you are rich enough not to need one.

Re:Tax deduction?

[FuturePastNow]

I'm looking forward to the first hydrogen hard drive.

Re:Tax deduction?

[eln]

Hydrogen is pie in the sky. The future of hard drives is switchgrass.

Re:Tax deduction?

[coaxeus]

I'm waiting until they come with a flux capacitor tbh.

Re:Tax deduction?

[the jalapeno]

Now I can park the heads for free!

Re:Tax deduction?

[fm6]

And I thought the first lame joke would be about car pool lanes. Terribly Californiacentric of me.

Re:Tax deduction?

[numbware]

Only if your hard drive gets over 30 MPG city.

Yes, but how many LOC?

[hoggoth]

> Seagate's pushbutton drive is capable of storing all of the following, combined: a 25-DVD movie collection, 15,000-song music collection, 15,00-photo image library, 50-hours worth of video, and 50 computer games, with 300GB left over

Bah, these measurements tell me nothing.
How many Libraries Of Congress can I store on this thing?! That's what I need to know!

Re:Yes, but how many LOC?

[kfg]

How many Libraries Of Congress can I store on this thing?! That's what I need to know!

With lossy compression one LOC per bit of available storage.

KFG

Re:Yes, but how many LOC?

[poogl3]

I agree with you in that I am tired of seeing these things as units of measurement. Would it be unreasonable to advertise just the capacity in GB (or GiB for those linux-minded folks)?

Re:Yes, but how many LOC?

[Pantero Blanco]

Seriously. 15,000 songs in midi format, 15,000 50x50 photos, 50 hours of low-bitrate video, and 50 text adventures don't take up much space at all.

I always wonder how they're counting the "DVD movies"...Raw and untranscoded? Transcoded to a 700MB avi? A direct copy of the DVD to your hard drive?

Re:Yes, but how many LOC?

[TheRaven64]

You can do that with lossless compression too, as long as you store the dictionary file (i.e. one copy of the LoC) somewhere else.

Re:Yes, but how many LOC?

[sharkey]

Likely, it's the legal correspondance regarding said DVDs with Warner Bros., FOX, et al and their lawyers that fill up the rest of the available space.

Re:Yes, but how many LOC?

[Anonymous Coward]

You can figure this stuff out yourself by doing math. You've heard of math, haven't you? It's that stuff you do with numbers.

Re:Yes, but how many LOC?

[Anonymous Coward]

15000X + 15000Y + 50Z + 50Q + 300 = 750

Sure, just solve for...Wait a second.

Re:Yes, but how many LOC?

[Anonymous Coward]

Storing dvd movies and songs on their hard drive? Quick! Call the MPAA! Call the RIAA! Someone must pay!

Re:Yes, but how many LOC?

[RobertLTux]

given that an IEEE LOC is 10 terabytes that would be 0.75 LOC (the drive is 750 gigs)

Re:Yes, but how many LOC?

[beren12]

Redundant? Can't we get a -1 Retarded mod?

Re:Yes, but how many LOC?

[Anonymous Coward]

(750 x 10^9) / (10 x 10^12) = 75 x 10^-3 = 0.075

Will it work?

[Umbral Blot]

Hibernation works by writing the contents of the RAM to the hard drive, so this would only work if you had = 256 MB RAM. I don't think too many new systems meet that requirement, and even less will after Vista comes out. Similarly if you want to save time on boot-up you would need to store all the necessary system files in that space, and few modern operating systems can cram themselves into that space.

Re:Will it work?

[ThisNukes4u]

But if you have 512mb of RAM, and even if only half of that is in the flash memory after hibernation, you're still saving ~ half the ammount of data that would otherwise have to be written and read from the disk, which is more likely than not a very substantial speedup and power savings versus no flash memory at all.

Re:Will it work?

[dorbabil]

There's no reason why just the core stuff the operating system needs to function couldn't be loaded into the flash portion, and the rest could overflow onto the HDD proper. I suppose that might require some changes to the OS.

On the other hand, open up task manager some time when you're about to go into hibernation. I think it's pretty unlikely that you've got more than a few hundred MB used when you're idling. I suppose if the whole reason why you're hibernating is so you don't have to reopen all of your web browsers and excel spreadsheets and visual studio IDEs, then you might be using a bit more. But I thought most people just used hibernation to shorten shutdown and start up times.

Re:Will it work?

[Short Circuit]

If the drive logic caches the most recent 256MB to flash, then that's 256MB that won't have to be read off the magnetic media. On a 768MB system, that's a third of the load time.

Re:Will it work?

[FireFury03]

When you hibernate, much of the stuff in memory can be dumped to the swap partition rather than to the "hibernate file". This means that on resume it can be swapped back in at a later time when it's actually needed rather than swapping it all in at once. So it's very likley that all the stuff that actually needs to be loaded immediately at resume time can fit into the flash memory.

What I want to know is what's the point in integrating the flash into the hard drive rather than just having it as an independent device that can be used how the software sees fit?

Re:Will it work?

[ryanov]

Part of the startup and shutdown time though is logging in and starting up all of your stuff. It's convenient not to have to do so because of hibernation (I know that I've been missing it since I switched to Linux -- only have just been tackling that now).

Re:Will it work?

[Umbral Blot]

But I, and soon Vista users, have 1GB of RAM, and I doubt that I am using 256 of it (I usually have a lot of apps open when I hibernate, as this is the point of hibernation, that you don't have to stop what you are doing when you want to turn off the computer).

Re:Will it work?

[HTL2001]

Exactly what I thought when I saw this. I think it might be better to have something like hybrid RAM if its just for doing hibernation, that way when you upgrade it still works. Hell, you can even save the whole state of your RAM from time to time if you want, to help recover from crashes.

It sounds like this would be very good though (as long as the size is enough to hibernate) so that when battery runs realy low the system has no chance of failing before the hibernation save is complete.

Re:Will it work?

[Jerf]

I don't know how they implemented this hybrid drive, and I don't know how much integration with the OS it does.

But I can say that if you did integrate with the OS, then you wouldn't necessarily need to cram all the system files into that space; you'd only need to cram all the system file pages that your system uses to boot. I have no idea how large this set is, except that it is bounded on the upper end by the total file size you thought would have to go in, but my intuition is that it's a lot smaller. (Windows "must" have IDE support in its system files to boot, but you wouldn't need the pages containing that code for a SCSI system.) I wouldn't be surprised either way that 256MB was sufficient.

In fact booting should already work this way in terms of what is actually read off of the disk (libraries are typically "mapped" and paged in, rather than being "loaded", for just this reason), so it's not even necessarily a lot of extra code, it's just that the hard drive would need to work with the system.

However, thinking about it this way points out a problem with the marketting spiel. The globally optimal behavior will be to treat it exactly as I highlight above; as basically another layer of swap between "in memory" and "on the physical hard disk", perhaps with a healthy dollop of space reserved for write-behind caching to catch the case where you save a file and don't want to spin up the harddrive for it. However, in that case, you'll want to use existing swap selection algorithms, which will eventually (and correctly!) flush the OS bits that aren't in continuous use out of the swap area. If you want to maintain those bits on purpose, you can special-case them in, but at the cost that you'll degrade the value of the flash memory whenever you aren't booting.

If we were talking 5GB of flash, I might be willing to make that trade, but at 256MB, I doubt I'd want to throw it away for that. Booting up may be annoying but it is still something you do fairly rarely.

Well, the more I think about it, the more interesting possibilities arise, but I gotta post this sometime. Anyhow, the upshot is that you definitely do not necessarily need to all of the files the OS uses to boot, just the bits it actually uses.

Re:Will it work?

[Anonymous Coward]

Well, I don't know anything about how windows' kernel works, but I wouldn't implement hibernation by dumping ram to the harddrive. I would implement hibernate by dumping DATA pages to the harddrive. All program code (ie pages marked read only) can be reloaded from disk -- no sense writing it.

Does window even implement pages like this? I'm pretty sure Linux and OSX do, but I wouldn't be surprised if windows did something different (ie stupid).

Re:Will it work?

[soleblaze]

What distro are you using? Hibernation works pretty much out of the box in Ubuntu Dapper (if it will work at all) and I believe it's also the case for FC5

Re:Will it work?

[ltwally]

"Similarly if you want to save time on boot-up you would need to store all the necessary system files in that space, and few modern operating systems can cram themselves into that space."

Not true. Flash memory is faster than hard drives (both in "seek times" and in raw transfer speed), so this would allow the core operating system files to be transfered quicker. As well, hard drives typically take a few seconds to "spin up" before being available to load data off of -- so this could also have the added benefit of being able to allow the OS to start booting while the hard drive is still spinning up.

Further, the 256mb model is only the beginning. This is the 1st generation product. In the future, we'll be able to purchase hard drives with 4+ gb of solid-state memory on-board, which will allow you to store pretty much all of your "core" OS files and such on it. Not even Windows Vista is going to be so large that the actually running OS couldn't fit on that.

Re:Will it work?

[HTH NE1]

But if you have 512mb of RAM,

you can store a little more than half a maybe.

Re:Will it work?

[brarrr]

I don't see why computers should write the RAM to swap/hd. Why not simply keep the ram powered and the data retained, but turn off all other system and sub-system components? Is there a technical limitation to doing this? I'm sure there is a break point where the time spent writing to disk uses less battery than powering the ram, but couldn't there then be two types of sleep?

Re:Will it work?

[ltwally]

"Hibernation works by writing the contents of the RAM to the hard drive, so this would only work if you had = 256 MB RAM. "

True, 256mb isn't enough to hibernate off of... however, what if the hibernation method were revised a little? If the OS were to clean out it memory caches (modern OS's cache just about everything -- disk, network, applications, etc.) and then only use as much hibernation space as is actually being used in memory? My workstation has 2 gigs of memory, but rarely am I actively using more than 512mb... so, theoretically, if hibernation mode were optimised for this technology, we could actually use this flash memory for hibernating.

Just an idea. Comes with its own advantages and drawbacks, like most things in life.

Re:Will it work?

[LnxAddct]

That is kind of like saying L2 cache is pointless because you can't fit 4 gigs of memory into it. Used wisely, this 256MB could be very useful.
Regards,
Steve

Re:Will it work?

[TomTheHand]

Yes! And we could call it... "standby", or maybe "sleep mode"! And people could use it when they want to boot up a little faster than hibernate, but they'd use hibernate if they want the system to use no power at all! Oh, wait...

Re:Will it work?

[ivan256]

What I want to know is what's the point in integrating the flash into the hard drive rather than just having it as an independent device that can be used how the software sees fit?

That requires software modification. As we know, most users are running either the current incarnation or the previous incarnation of Microsoft Windows. A change to Windows that would use such a device would be two versions out, which means three PC lifecycles before said seperate flash device has any signifigant market share.

In other words, they made it this way so they could sell them now.

Re:Will it work?

[generic-man]

Windows XP already gives you the choice* of standby and hibernate. Standby keeps the memory and other components powered at a minimal level. Hibernate writes the contents of RAM to disk and shuts the system off. The difference is in battery life. Standby consumes some small amount of power; hibernate consumes virtually none. You can't swap a fresh battery in while your laptop is on standby (unless you're plugged into AC power or you have two battery slots) but you can while your laptop is hibernating.

* as long as you've turned on Hibernation in Control Panel > System

Re:Will it work?

[Stellian]

Also, the best part of the memory of a 512MB OR 1GB Laptop is free under light usage like a word processor and Winamp. Well, not actually free, any free memory is used as disk cache for reading, but that can be discarded.
Also, many of the memory pages are mapped as "read-only", for example all executable files running. Those do not need to be swapped out or written to the hibernate file - they can be discarded and read back again from the hard-drive when the thread executing them becomes active.

As an extra idea to speed up hibernation - what about compressing the read/write pages with a fast algorithm like lzop prior to writing them to disk? I regularly work with core dumps and they are usually compressible with gzip at a ratio of 15:1 or better. Here's what the homepage for lzop claims:

On modern systems, when making backups of terrabyte of data, lzop is usually IO-bound and not CPU-bound, which means that you can both decrease storage requirements and effectively reduce backup time by quite an amount.

Re:Will it work?

[Ender Ryan]

What I want to know is what's the point in integrating the flash into the hard drive rather than just having it as an independent device that can be used how the software sees fit?

Compatibility and ease of integrating into existing systems I'd assume.

Re:Will it work?

[rbanffy]

You can compress the memory dump so it takes up less space. If you achieve a 2 to 1 ratio, you can cram 512 megs of RAM into 256 megs of flash. You also invalidate any buffers and caches you may have active so that only programs and data is dumped.

If Munin is right about my computer, there are 341 megs serving as caches and 51 megs used as buffers. If I am reading the graph correctly, I am using just over 400 megs for programs and working data. I am not touching the swap file.

Another optimization would be to give more importance to foreground programs than to servers and restart them after you can safely restart the more interactive parts.

Of course, if Linux can't do this, these drives are useless to me.

Re:Will it work?

[Anonymous Coward]

You sir, are a fucking moron. You're completely missing the point.

Re:Will it work?

[armitage_23]

If you'd read the article, you'd know that the non-volitile memory only works with Windows Vista.

On 2000 or XP, the drive will act like a normal drive, albeit with more cache.

Re:Will it work?

[Anonymous Coward]

You do not have to swap out everything before hibernating. Things not to swap out include:

• disk cache
• any block that is already on the disk, such as the kernel image or images of running executables
• free blocks

Also, you do not have to swap in everything at wakeup. You can (and probably should) use demand-paging, reading only those pages that are actually needed.

That said, 256 MB seems a bit on the small side to me, too. I would guesstimate that you would start seeing benefits if you had a Flash backing store that is about 50% the size of physical memory.

Re:Will it work?

[Heembo]

> this would only work if you had = 256 MB RAM. I don't think too many new systems meet that requirement

I live in a world where most computers around me have at least 1 gig of ram these days. Even Dell's bottom feeder computers have 512 base with frequent 1 gig upgrades for free....

Re:Will it work?

[FireFury03]

Compatibility and ease of integrating into existing systems I'd assume.

Compatability really doesn't count since it requires specific support in Vista, so they could've implemented it to use a completely separate flash device rather than one that's built into a drive.

I'm taking the "ease of integration" thing with a large pinch of salt too - if this is primarily for notebooks then you can more or less discount upgrades (how many people really upgrade the drive in their notebook?). So this will be installed by the notebook manufacturer - i.e. there's no real reason for putting it in the drive rather than on the motherboard. And it seems like it might be more sensible to hang the flash chips off the north bridge directly rather than on an ATA bus since it would probably improve read speeds.

Re:Will it work?

[SanityInAnarchy]

So what?

Windows doesn't know how to do software RAID either, but Nvidia's drivers and BIOS provide that. Yes, it sucks that Windows is proprietary, but if you went to MS and said "We'd like to improve your OS for you", I think they'd jump at the chance, and they'd help you (help them) any way they could.

Re:Will it work?

[Kadin2048]

if you went to MS and said "We'd like to improve your OS for you", I think they'd jump at the chance, and they'd help you (help them) any way they could.

Do you think they'd still be game if I wanted to improve it by imprisoning all of their programmers and dumping the source code in the ocean?

Oh...you mean they have to think it's an improvement, too? Crud.

Re:Will it work?

[Jokerz17]

you can store a little more than half a maybe.

But what if I want to store a whole maybe?

Re:Will it work?

[Nikker]

Don't forget data in RAM is seldomly compressed. Being in RAM it is in the ideal location to be compressed especially with todays processors the south bus would likely still be just as saturated as RAM -> HDD.

Re:Will it work?

[ChatHuant]

Windows doesn't know how to do software RAID either

You mean like that?

Re:Will it work?

[brarrr]

Ah - I'm a mac guy, so no joy for me... for now. Though I want to say that the memory remains powered when in sleep mode even though the battery can be swapped out. There is a battery that holds things for 10 or 20 seconds in that case, so no differentiation between standby and hibernate.

No windows laptop- -no reason to have a difference.

Re:Will it work?

[ivan256]

Care to mention where it says that? The only thing I saw in there about being Vista only is the support of 4k block sizes.

Re:Will it work?

[generic-man]

I heard that was true so I tried it with my 12" Powerbook. I put the Powerbook to sleep, took out the battery, and popped the battery back in 3 seconds later. The whole system went dead and I had to boot up all over again.

Apparently the newer Mac laptops hibernate by deafult and you can mess with the standby/hibernate configuration with command-line tools. Do so at your own peril.

Re:Will it work?

[ivan256]

Virtualizing devices is something that is easy to do in a device driver.

Tasks that require knowledge of what data means without cooperation from the software generating the IO are difficult or impossible to do in a device driver depending on the task. It would be hard, where hard is a relative term in the context of software raid being easy, to accelerate hibernation in a block device driver. It would be impossible to do it well.

Yes, I write storage device drivers for a living, and have personally implemented software RAID and various types of virtualization and multi-pathing in device drivers.

Re:Will it work?

[47Ronin]

Yes, this type of technology has already been implemented, if not in flash form. Take any Mac laptop from the last, oh five or six years. Close the screen, and within a second it goes to sleep. You'll see the the front LED pulse which tells you it's in deep hibernate mode and everything's suspended.

Now, open the screen. Within a second, the screen is back on and you're back to whatever you were doing, your network is reconnected, and you're ready to work.

It doesn't require any fancy hardware. Apple has this working in Mac OS X on both PowerPC and Intel systems. Funny thing, this instant sleep/wake feature is available on Apple's laptop AND desktop systems.

Re:Will it work?

[ChrisA90278]

Everything in RAM does not have to be saved. If you look at a computer with 1GB of RAM you see that much of what is in RAM is just a disk buffer cache and pages of executable code that were paged in. This stuff can simply be trashed as it will be paged back in after the machine wakes up. Saving LESS to RAm can actually make start up FASTER because less data neds to move between flash and RAM.

Re:Will it work?

[mosdhorton]

The last time I saw a discussion on these flash/magnetic "Hybrid" drives, the primary motivation for their design was power savings - performance enhancement was sort of a bonus. The focus wasn't in improving 'Hibernation' since that really isn't a power-saving feature (IMO) - it's more like a bookmark function. On the other hand, if the cache was sufficient to allow the laptop to "Wake" from "Sleep" mode without having to spin the disk - that would probably be a significant power savings.

Mainly, though - the power savings was realized in the number of times the laptop was able to extract some oddball bit of hard-disk information - say a page from cache or a 2k cached web page - without having to spin up the disk for it. I also seem to recall that the hybrid disks included some additional ATA commands that would allow updating OS drivers to manage that cache more intelligently - although that concept may be a figment of my mediocre memory.

Anyway - that's what I recall from a few months ago in 'Popular Science' or some other such sort of thing.

Re:Will it work?

[ryanov]

Dapper. It actually works pretty well. I've found that I've had more consistent luck since I switched to nvidia-glx (this is a new-to-me laptop at the same time as the Dapper "upgrade" so I'm not sure what is broken how yet). Suspend to RAM doesn't work at all with nvidia-glx (but did work somewhat with the nv driver, but it was unpredictable).

too small

[Blissett Luther]

Only 256 MB? It's too small for a memory dump even on a notebook. I guess it's not "upgradable" so what's the point?

Re:too small

[Anonymous Coward]

It's too small for a memory dump even on a notebook

Yes, it's too small to dump all of the memory for a 512KiB or 1GiB laptop. That is unnecessary if the goal is to resume quickly. The only part of RAM that must be perserved for rapid resume is the 'working set'; that usually small part of process memory that is paged into RAM to actually execute programs. The rest (what appears as 'virtual' memory size in Windows) can be safely discarded or saved to disk.

I own several 1GiB+ laptops. The 'working set' size under usual conditions would fit comfortably in 256MiB.

Re:too small

[cthulhu11]

It's called compression. Look into it.

Any benefit to existing laptops?

[WoTG]

The PR blurb is a little light on the details. Does anyone know if there will be speed benefits (or, IMHO, less likely power benefits) for existing laptops? I.e. should I look forward to giving my laptop a bit of a boost with one of these drives? I know that Vista is supposed to have a lot of code to really benefit from hybrid drives... but I imagine that at least some benefits might be available to XP or Linux.

Does anyone smarter than me know more about these drives?

Re:Mbytes?

[hoggoth]

> How about Megabs?

Are those 6 minute MegaAbs?

What about maximum read/writes for flash?

[PrescriptionWarning]

don't some flash memories have limited read/writes far below platter hard drives? something like in the range of a couple thousand I thought... does this mean your hard drive will die even quicker, or will they make these drives eventually have removable and replacable flash cards, such as SD or something?

Re:What about maximum read/writes for flash?

[flooey]

don't some flash memories have limited read/writes far below platter hard drives? something like in the range of a couple thousand I thought... does this mean your hard drive will die even quicker, or will they make these drives eventually have removable and replacable flash cards, such as SD or something?

It's unlimited reads, but limited writes, so assuming you're using it to store OS code, the limited writes probably won't be a major problem. The limitation is usually in the low millions as well.

Re:What about maximum read/writes for flash?

[bishiraver]

Most commercially available flash memory has a limitation of about 1 million rewrites. A better solution to this, rather than putting the flash memory in the hard drive: Put a flash card, removable and upgradable similar to how RAM is on a laptop, and mount it as your Suspend To Disk partition. Except, well, that doesn't work well on windows, I don't think... Then again, I've never used Windows on a laptop. Something else that would benefit from Flash rather than hard drive, is virtual memory.... reduce seek time, allow read from flash at the same time as read from hard drive, etc. Would significantly improve the performance of virtual memory.

Re:What about maximum read/writes for flash?

[Fallingcow]

Provided that one could get a flash drive to show up as a regular IDE drive in BIOS, could one not mount /root or /lib/modules/ there to greatly speed up boot times without the hassle of making a bootable USB flash stick?

Re:What about maximum read/writes for flash?

[cyngus]

Its true that there are a limited number of writes, but its also not necessarily a fatal probelm. As bits in the flash memory fail, you can mark them as unavailable. This will shrink your solid state cache over time, but it will allow you to keep working until the physical disk fails. Very graceful degredation.

Re:What about maximum read/writes for flash?

[smbarbour]

Considering that an adapter exists to connect CompactFlash media to an IDE interface cable, I'd say that you can get a flash drive to show up as a regular IDE drive. There are even existing products that do the same thing prepackaged as IDE devices. To save you some time, here is a link to the Google search: Keywords: Non-Volatile Solid State IDE Drive

Re:What about maximum read/writes for flash?

[TheRaven64]

I would use this space as a write cache. When the hard drive is spun down, you write the data to the flash instead of to the disk. When the flash is full, you spin up the drive, do a 256MB write, and spin it down again. Operating systems tend to be quite good at read caching, but spinning up the drive for small writes really hurts battery life a lot.

Re:What about maximum read/writes for flash?

[epiphani]

assuming you're using it to store OS code

Thats exactly what I want to do with it. In fact, thats all I want the system to do AT ALL. Hybrid drives are nice and all, but why isnt there one damn SATA/EIDE/ATA/whatever 4GB+ flash-only drive?

This would be invaluble to us in a server environment. All our storage is remote already, for gods' sake just give me a puny little solid-state drive that will rarely fail. Drives are the biggest thing we loose. It would drop costs for us all over the place.

I'd even use a bigger one at home. A pair of 6GB flash chips would be plenty for me and my desktop. I'll put all my storage on one big machine, and hide it in the basement.

Why isnt anyone selling this?

Re:What about maximum read/writes for flash?

[creepynut]

That's not an issue with the BIOS not recognizing the Flash drive, that's an issue with the BIOS seeing it as a USB device, rather than a regular IDE or SATA device.

Some BIOS's are more than cabable of booting from a USB device, so who's saying you couldn't do this already?

Re:What about maximum read/writes for flash?

[Klaruz]

First google result: http://www.adtron.com/products/A25fb-SerialATAFlas hDisk.html

Flash drives have been used in embeded stuff for a while now.

Re:What about maximum read/writes for flash?

[scrod]

Why isnt anyone selling this?

They are.

Re:What about maximum read/writes for flash?

[LnxAddct]

No, for years now flash memory from every major vendor has been certified with a *minium* of one million writes, which is more than competitive with your standard harddrive.
Regards,
Stev

Re:What about maximum read/writes for flash?

[Anonymous Coward]

Yeah buddy, the designers totally didn't think of that. Good thing you took a break from your helpdesk job to post that VALUABLE info on this hunk o' shit website.... I hate Slashdot and everyone who loves it.

Re:What about maximum read/writes for flash?

[Fallingcow]

That would be why I said, "Provided that one could get a flash drive to show up as a regular IDE drive in BIOS [....]" and "[...] without the hassle of making a bootable USB flash stick" :)

Re:What about maximum read/writes for flash?

[Nikker]

That would work if the data written was always smaller than the cache, most notebooks for example have a hibernation file identical to the size of availible RAM. With 256MB cache each write would likely completely fill your memory, thus all or most sectors would die simultaneously.

Re:What about maximum read/writes for flash?

[cyngus]

There are two problems with your logic. One is that they aren't going to use this hard drive flash cache just for hibernation, but rather as a general purpose cache, probably controlled by software on the drive itself. Secondly, even if it were used for hibernation, sectors would fail at different times due to manufacturing variations between them, although you'd probably expect something like a normal distribution curve of failures.

Re:What about maximum read/writes for flash?

[Nikker]

The only point I was trying to make is GP said there would be a gradual failure as each bit reaches its lifetime expection(ie, 1Mil cycles etc). This is fine for end user data storage since most data written has more significance and is written and erased less frequently(possibly tens of times a day). Since the goal of the controlling logic would aim for highest effency of data transfer the drive has the capability of reading and erasing possibly a thousand or more times per hour. With data stacking techniques to keep currently "popular" data resident new data would be consistenly filling all the availible bits. In short it works out to be the same as mounting your /tmp directory onto your flash drive.

Flash drives are awesome as floppy disks because they are quick to read and write but will fail quickly if used the way a harddrive could fill and empty it on such a tight read and write cycle.

Old objects of lust

[also-rr]

This brings back a memory of a very, very, very long time ago when I was fortunate enough to get to touch a computer that had its root filesystem on a 250mb solid state disk, so that it only had to touch the much slower mechanical drives infrequently. For it's day the thing was a monster with speed that made my own systems seem inadequate in every way. So what did we do with all of that raw, untamed power? Played nethack.

Re:Old objects of lust

[radish]

Sounds like my PDA....

lifetime of flash?

[Chirs]

I'm a bit worried about how long that flash memory is going to last. It's got a limited number of write cycles, and presumably everything going to the drive goes through the flash cache.

Re:lifetime of flash?

[lawpoop]

"presumably everything going to the drive goes through the flash cache."

Doubtful, given the wear-and-tear issue you point out.

I rather think the flash cache is to store the RAM data when the machine goes into hibernation, and to load back in quickly when the machine comes out of hibernation. Without flash, you would have to wait for the hard drive to spin up before it can retrieve the RAM store.

Re:lifetime of flash?

[fredistheking]

With defect management it shouldn't matter. There is no way the average number of writes per sector is going to get anywhere near the limit. If a particular sector is getting close, simply switch its address with one that isn't used very often.

Since Seagate is already defect managing the disk with their firmware, I don't see it being a big challange to have it defect manage the flash as well.

Re:lifetime of flash?

[athakur999]

The write speed of the fastest flash cards isn't that much different from a good hard drive. It doesn't make too much sense to use it as a cache for this purpose. However, since I can't RTFA I don't know if they're using a different type of flash memory than what is common used on flash memory cards.

Re:lifetime of flash?

[archen]

Well assuming Seagate isn't stupid, they would probably put more robust flash into the drive, so I'd assume about 100,000 writes. Assuming you did a hybernation every 10 minutes all day, it would last about 694 days. That's pretty extreme and still lasting nearly two years. With average use I'd say it wouldn't even be a factor for most people.

Re:lifetime of flash?

[maxume]

So you thought of it, but you don't think that people who make hard drives for a living are going to think of it? Really?

It would seem much more likely that the drive looks for files that are read often and stores them in the flash.

Re:lifetime of flash?

[jellomizer]

Why?
If it dies it will work like a normal Hard Drive. You only have things to gain while it works. And when in so many years (for most usage well after the Computer is obsolete) It will run at the speed of a normal drive.

It may not be good for Servers and espectially Real Time Computers. But it is like saying to a person who is gives you $10,000 with no catch, Why should I take the $10,000 Ill just spend it in time, after that I wont have it, I will only accecpt free money when I can live on the interest.

Re:lifetime of flash?

[Billly Gates]

Which flashdrives use this?

My old 512 that is a year and a half old is taking a long time accessing back and forth but I have not lost any data. Yes I make sure its defragmented. I wonder if defect management is working in the firmware on it as it appeared strange.

Guess its time to buy a new one

Re:lifetime of flash?

[fredistheking]

There isn't any defect management on a flash drive. I don't believe they have any firmware at all. On a disk drive, however, the flash could be defect managed since the firmware code already exists to defect manage the disk.

software side

[F�an�ro]

This seems to be something that could be done just as well on the software side by combining a harddrive and a flash card reader.

Are there drivers that can do this?

Re:software side

[Anonymous Coward]

Flash cards are slower than most hard drives. This memory is much more similar to RAM.

Re:software side

[Anonymous Coward]

I don't know of any that will work on existing system; however, Windows Vista is supported to allow this with thumbdrives.

Re:software side

[Siker]

It probably could but it would be a very dangerous solution. If the flash is on the harddrive, the harddrive can ensure the integrity of the data. If you write a MB of data to the harddrive, you might not know if it has been cached on the flash or if it is actually on the disk, but at least you know for sure it's on the drive. If your operating system's disk driver crashes the next second, you're fine anyhow.

Re:software side

[Adelec Bakkal]

Flash cards are slower than most hard drives. This memory is much more similar to RAM.

Flash HDD && Flash ~ RAM ? I thought RAMs were much faster than HDDs...

Call me a cretin, but...

[one-eye-johnson]

What's the difference between a 'hybrid' drive and a drive with a really big cache?

Re:Call me a cretin, but...

[XMyth]

Cretin. :)

The flash on hybrid hard drives is used to store data (say a copy of system RAM when hibernating) after the machine is off...think solid state storage like a USB drive, not solid state like system RAM.

Re:Call me a cretin, but...

[TeknoHog]

What's the difference between a 'hybrid' drive and a drive with a really big cache?

The article highlights faster resume times from hibernation. In that case the power has been off, which would empty the cache.

Re:Call me a cretin, but...

[99BottlesOfBeerInMyF]

What's the difference between a 'hybrid' drive and a drive with a really big cache?

Cache is volatile, flash memory in a hybrid drive isn't. Thus a hybrid drive could save time when you boot, while a large cache won't.

Re:Call me a cretin, but...

[Tony Hoyle]

Hybrid drives have a petrol engine that starts up whenever you need to actually use them, and draw power the rest of the time. :)

Re:Call me a cretin, but...

[timeOday]

In addition to what everybody else said, Flash is very cheap, so you can have it in quantities beyond normal Cache sizes. The advantage of that is you don't have to spin up the hard disk itself so often.

Death of Harddrives?

[Billosaur]

The Momentus 5400 PSD is Seagate's first hybrid hard drive, incorporating 256 Mbytes of flash memory that serves as a fast cache for booting and saving data. When booting the PC, the operating system loads data from the flash memory first, speeding bootup times and negating the need to quickly spin up the drive, a power-consuming process.

Given the rapid pace of development of flash memory, how long until hard drives are gone altogether? It would seem the breakout of flash memory in the marketplace is bringing us one step closer to relaible instant-on systems, with none of the tedious waiting for drives to spin up.

Re:Death of Harddrives?

[flooey]

Given the rapid pace of development of flash memory, how long until hard drives are gone altogether? It would seem the breakout of flash memory in the marketplace is bringing us one step closer to relaible instant-on systems, with none of the tedious waiting for drives to spin up.

I'd imagine that hard drives will go away only once they find something akin to flash that isn't limited in the number of writes. Having a limit of a million writes is completely reasonable for iPods, cameras, and other devices where you do infrequent large writes. Having /tmp, home directories, or so forth on flash memory could burn it out pretty fast, though.

Having a flash device for the OS and programs and a hard drive for general purpose storage, though, that I could see being feasible in not too long.

Re:Death of Harddrives?

[vijayiyer]

Hard drives aren't going away soon, the same way that tapes aren't. Instant on isn't necessary for a lot of people, a lot of people never boot or shutdown their machines. Flash will instead take a certain segment of the hard drive market, where portability and power draw are more important than capacity.

Re:Death of Harddrives?

[Surt]

Some flash is up to about 3 million writes already. At 10 million writes the problem is effectively solved, they'll be able to warranty their flash for continuous writes for about 5 years at that point, matching the warranty on your hard drive.

The write limit is not going to be the barrier to replacing hard drives for nearly as long as price and size are going to be.

Re:Death of Harddrives?

[geobeck]

...relaible instant-on systems, with none of the tedious waiting for drives to spin up.

Good question, but not quite the right one. Most of the time you wait for your computer isn't the hard drive spinning up, but the OS transferring information from the hard drive to RAM (and to other parts of the hard drive). Flash RAM will do this faster because it won't have to wait for the HD's heads to access a particular part of the spinning drive, but it will still take time.

We'll never have the same instant-on response that we had with our Commodore 64's!

Re:Death of Harddrives?

[Gat0r30y]

Probably just the opposite. Right now, hard drives are being made in the 1.8 in form factor, and Toshiba has a .85 in form factor drive. Since hard drives are so much cheaper per GB, and the price is constantly falling (much better competition in the HDD market than in flash) you will quite likely have a hard drive in your computer for a good long while. Will software manufacturers eventualy switch to a flash OS? That sounds more reasonable, and especially good for them from a licence management point of view.

Re:Death of Harddrives?

[John Courtland]

10 million writes, writing once every 5 seconds, not accounting for defect balancing, leaves you at about 578 days. If the disk is constantly full, this will be a problem, unless it's cheap and easily hot swappable, or if the disk is under 50% utilization for its life. I'd still probably keep logs on a regular magnetic platter-based drive.

Re:Death of Harddrives?

[Surt]

You have to factor in defect balancing because you can't come anywhere close to writing the entire memory every 5 seconds. It takes closer to 30 minutes.

Re:Death of Harddrives?

[CryoPenguin]

Flash write cycles are limited per block, not for the whole drive. Do you overwrite your whole harddrive once every 5 seconds? It's not even capable of writing that fast. Consider a 100GB drive, writing continuously at 50MB/s for 5 years. With perfect defect management, that would only use up 80,000 write cycles.

Re:Death of Harddrives?

[Peldor]

When you can get 300GB of flash memory for $100, maybe conventional harddrives will start to die out.

As it stands now, you're looking at $24/GB for flash media vs $0.33/GB for magnetic. You can nitpick the numbers, but it's currently close to 100x more expensive for flash.

Re:Death of Harddrives?

[sploxx]

Some flash is up to about 3 million writes already. At 10 million writes the problem is effectively solved, they'll be able to warranty their flash for continuous writes for about 5 years at that point, matching the warranty on your hard drive.

And you can wear it out evenly with specially designed file systems and/or storage patterns - which gives you 250MB*a million cycles sum data transfer to the flash and not only a million cycles for each single location.

Re:Death of Harddrives?

[Surt]

Exactly right, and in fact they don't even need special file systems, they have load balancing logic on board the drive to even out the write patterns, so that it won't matter what file system you use.

Re:Death of Harddrives?

[david.emery]

What about paging/swapping? That's the primary thing that happens on rotating media that would bother me with a 10m duty cycle over 5 years.

And from a system perspective, where would you put the 'smarts' to continually move around "known system files" such as the pagefile, or the inode file? Should that be done on the controller, within the memory itself, or back on the OS kernel?

I hope this does not become an OS Kernel responsibility.

        dave

Re:Death of Harddrives?

[Billly Gates]

Do you have any idea how much swapping Windows braindead vm uses?

Even if you have plenty of ram it constantly swaps. It would eat flash ram for breakfast. Windows sucks balls in this area big time.

I am amazed at how much longer my notebook lasts under Linux because its not constantly swapping unlike XP. The hard drive light is always on under XP and it drives me nuts.

Re:Death of Harddrives?

[Surt]

It doesn't really matter. Flash is fast approaching the point where writing at the maximum rate (say 60MBytes/sec) for 5 years straight won't use up the write capacity. It doesn't matter what your workload looks like, you won't be able to use up all of the writes in any reasonable time frame (unless the write speed goes way up, but that doesn't look likely).

I'm not surprised

[Anonymous Coward]

I've been playing with Damn Small Linux using a 633 MHz pentium motherboard (attic-ware) with a 12 volt power supply and a 256 MB flash card. It uses an average of 1.5 amps. (monitor not included) When my ancient Thinkpad is accessing the hard drive, it draws about 4 amps. Some of the current is driving the LCD but my guess is that when the hard drive is being used, it soaks up about half the power. If you could avoid using the hard drive, you could just about double your battery life compared to what you would get if you were using it all the time.

Having said the above, it occurs to me that you could use some of the techniques on a regular laptop that Damn Small Linux (DSL) uses. Flash memory can only be written to a finite number of times. In order not to kill the flash memory, DSL runs entirely in memory. (If you want to write to the flash memory, you have to explicitly mount it.) So, if you were to tailor your operating system to avoid using the hard drive the same way DSL avoids using the flash, you should be able to significantly increase your battery life without special hardware.

Re:I'm not surprised

[2300cc]

Loading the OS into ram is done by creating a RAMDISK filesystem which I believe is universal among most Linux distributions. If anyone's interested here's a link on how to do it: http://silent.gumph.org/content/4/1/011-linux-on-c f.html It's for compact flash but I'm sure it could be adapted to work on a USB stick without too much trouble.

Boot speed with a RAMDISK setup is not too great. During boot the kernel is loaded from the flash, a compressed filesystem is copied into ram from the flash and then that gets decompressed. The fastest I could get my 300Mhz PC104 system to boot to a console prompt was 67 seconds (and that's with a stripped out kernel and barely anything running on boot...maybe 10 processes). Also keep in mind that your main filesystem can only be about 2/3rds the size of memory or else you'll run out of room when it goes to decompress the RAMDISK filesystem (for example, an 80mb decompressed filesystem and 30mb compressed filesystem would be the max for a system with 128MB of ram).

flash memory lifetime?

[techmuse]

It used to be that flash memory only worked reliably for a limited number of write cycles. Is this still the case. If not, will this greatly limit the life span of these drives?

Re:flash memory lifetime?

[swv3752]

Old flash like that in my Zaurus SL550 PDA and older Compact Flash cards have ~100,000 writes. Newer Flash has something on the order of ~1,000,000 writes. As it is used to cache seldom updated things like OS files or hibernate files, it shouldn't be an issue. Mind you, it should have been 1-4Gb instead of 256mb, so that the OS and other useful things could be cached entirely.

Re:flash memory lifetime?

[Billly Gates]

My ram stick is taking longer and longer to access and I have a hunch its flash ram is getting burned out.

At the lab computers at school I use my flashdrive to run javadocs, javac, and gvim as well as my files. a million writes can easily be obtained by large apps such as the 70 megs of html of javadocs.

Mirror?

[dogmatixpsych]

Anyone happen to get a mirror of the links? They are completely dead at this moment.

Re:Mirror?

[Anonymous Coward]

well that's what happens to a server when it uses flash memory

Does this matter when you have a smart OS?

[mary_will_grow]

If you look at "top" closely, you'll see even if only half of your ram is stuffed with porn and chat programs, the kernel is still making use of that remaining RAM. It would be moronic to just leave RAM sitting unoccupied. A lot of it is used for IO buffering, including your hard drive. So why not just use this mechanism? Why is it, from an IO-buffering-OS-user's perspectiive, any different having that info sitting in flash on the hard drive, instead of in your ram?

OK I guess I can think of a few reasons...

The flash wont need refresh cycles to keep its data intact, so that gives you a power reduction...
The flash can still retain its state even when you shut down, so "wakeups" should be faster..
The hard drive is in charge of the caching, taking some thinky think load off of the CPU.

but from a performance perspective, it seems that Linux would do better with 256MB of faster, closer, shinier RAM instead of a wad of flash.. Plus your caching mechanism can be improved without having to buy a new hard drive.

Re:Does this matter when you have a smart OS?

[Zemplar]

"Does this matter when you have a smart OS?"

You missed the point, this idea was introduced at the "2004 Windows Hardware Engineering Conference".

Perhaps Microsoft will use this memory space to load such advanced technologies as anti-virus, anti-spyware, defrag, or other useful novel ideas...at least to the Windows-world?

Basically what this boils down to is if you have dumb software, you need smart hardware.

Re:Does this matter when you have a smart OS?

[TheRaven64]

Write caching. You don't want to leave data in your OS cache unwritten for very long. If the system crashes (hardware or software failure) then all of that data will be lost. If you start writing data while the drive is spun down, then it has to spin up (lots of power) and then write it, then spin down a bit later (using a fair bit of power all the time it is spinning). This design could, conceivably, write the little trickles of data (4K here, 16K there) that would otherwise keep the disk spun-up to the flash in a linear fashion. Once the flash is full, it could write the entire block to disk at once, and then spin down. Combined with good read caching in the OS, this could dramatically reduce the amount of power the disk uses in normal situations.

Of course, I have no idea if this is what it actually does. It's what I'd do though.

OS support?

[madnuke]

Are there any OS's that support this yet, the only one I can think of is Vista but its only in Beta 2 so might not have it supported yet. Still great idea, move away from magnetic storage lets have everything flash.

Re:OS support?

[Overzeetop]

I got the impression (via the summary - I never read the articles, of course) that this was internal to the HD, and therefore would be driver independent.

Me? I've got 2GB of RAM, so it won't speed up hibernation for me. Still, getting another half hour or more out of my laptop would be nice.

Re:OS support?

[Overzeetop]

Okay, so I finally clicked through to the article referenced by TFA, and yes, most of the good features (not necessarily ennumerated) are only in Vista. Sux.

MirrorDot to the rescue!

[PMJ2kx]

PC Mag:
http://mirrordot.org/stories/838dd483f468b1c95ac05 27291fe59b7/index.html

Extreme Tech:
http://mirrordot.org/stories/c6b3da4e4e2b800ddf83f 2d492109dd1/index.html

Regarding the other announcement, DB35 series...

[Anonymous Coward]

They also launched the DB35 series, supposed to be optimised for DVRs - quiet acoustics, capacities from 80GB-750GB, optimised for sequential streaming (and apparently up to ten simultaneous streams), long-haul reliability.

I might want to check those out for personal storage too. It sounds like they might make a nice, quiet, fileserver for my home, with the right case (I was thinking P180) and components.

There's this interesting snippet, though, which concerns me, in the DB35 series' product datasheet (PDF, 2 pages, 122KB):

"Drive security tools enhance fair use of digital programming by helping manufacturers implement appropriate digital rights management technologies."

(To give context, the `manufacturers' it is referring to are DVR manufacturers, which in my case, would of course be me. Maybe I should try MythTV.)

I am, of course, one of those people that feels the only appropriate Digital Restriction Mechanism is none at all... does anyone, anywhere have the faintest idea what they're going on about with that? What on earth has a hard disk got to do with DRM? (In the Vista Home Premium/Media Center/East Fork/"ViiV" stuff they might mean when they say that stuff, it'll all be encrypted before hitting the hard disk anyway, because it's a form of the WM-DRM, and wouldn't be allowed unencrypted across the SATA/PATA bus, so it's none of the hard disk's business there either...)

Is that, perhaps, pure marketing fluff that means "You can password-lock or encrypt the drive", or something more sinister? Anyone know? (And you've gotta love the way they justify it by using the phrase "enhance fair use", which is of course, the exact opposite of what any DRM is designed to allow.)

Flashy Mobiles

[Doc Ruby]

I've got an Dell notebook with only 10GB (IDE) HD. I'd love to replace it with Flash cards. They're about $45:2GB up to 4GB, in multiple formats. A bank of CF/SDIO/USB slots, or just an IDE/whatever adapter, plus the cards, would fit inside the current drive's slot. And offer much better power, weight and heat loads. With hotswappable filesystems, upgradeable in small chunks and pluggable into other devices, carryable in pockets.

I don't see how <20GB HDs have any place in the portable market anymore (outside of tiny niche multimedia producers), as even $35 80GB HDs are overkill for most people who network, as most everyone does. If every notebook, handheld, iPod, phone and other mobile device used Flash instead of HDs, Flash prices at that industry scale would drop, capacities would multiply, and $5:GB up to 32 or 64GB would be common. While much of the rest of the cost of the device would be lower without extreme measures to accommodate the hungry, inefficient HD.

Re:Flashy Mobiles

[WhiteWolf666]

If you want a Flash drive..... buy one ;-)

http://www.bitmicro.com/products_edisk_25_ide.php

I'm not 100% certain where you can purchase them, but when I looked into it ~6 months ago I did find some avaliable for online ordering.

Re:Flashy Mobiles

[Doc Ruby]

Those high-capacity industrial drives are way overkill for what I was talking about. They cost way more than the $20:GB Flash cards I mentioned, and aren't available in small capacities (and therefore low prices). And their capacities are large enough that they don't save enough on weight or power.

Let me know when you find a simple IDE/Flash adapter that I can plug the cheap commodity Flash stuff into, to replace my 10GB IDE HD.

Re:Flashy Mobiles

[Anonymous Coward]

Direct CF to IDE 2.5" adapter, $32 USD
PQI 8GB 100x CF card, $173 USD ($21 per GB)

10GB is an odd capacity for flash mem.
-e

Re:Flashy Mobiles

[WhiteWolf666]

Well, here are three posibilities. None are perfect, but I'm guessing they are workable:

1. Desktop usage:
http://www.monoprice.com/products/search.asp?keywo rd=2105
And a 2 or 4 port PATA raid card,
or
http://www.topmicrousa.com/st-123cf.html
and a SATA capable motherboard or SATA raid card.

The second has the advantage of being easily hotswappable.

Disadvantage: Not as cheap as you might like. Probably $50-$100 investment required.

2. Notebook usage:
http://americanesuperstore.stores.yahoo.net/cfad-0 03.html
Get two. Assumes your notebook has 2 PCMCIA slots, and you don't mind wasting them both.

Advantage? Internal
Disadvantage? No open PCMCIA slots. Most modern notebooks (unlike my Inspiron 8200) only have 1. Notebook must support booting from PCMCIA IDE

3. Notebook usage:
http://www.lexar.com/readers/pro_reader.html
Lexar daisy chainnable USB/Fireware CF reader.

Advantage? Does exactly what you want.
Disadvantage? Probably pricey. (~$80) External. Notebook must support booting from USB/Firewire.

Other items of note:
http://www.sprysoft.com/card-reader-internal/p_288 29.html

This allows you to plug in a combination of flash devices. SD, CF, whatever else. You could probably get up to 10 or so GB. Requires a 5.25" bay.

http://www.lexar.com/ufc/index.html

If you are really handy with wiring inside your notebook, you could probably stack these tiny Lexar USB Flashcards. They are slightly larger than a USB port; pretty damn small, in other words.

I guess I misunderstood you the first time (in terms of price/convenience), but I know the tech is out there. A couple guys with some basic soldering capabilities could probably through your dream, 2.5" IDE/Flash adapter together in a matter of days, with either using the IDE connection, or slapping together something via internal USB. Either way, what you are asking for is probably only not avaliable because of lack of market demand.

Re:Flashy Mobiles

[ygthb]

Recipe for long couch surfing.

Z505*** - ~$50-$100(ebay)

3x battery - $100

CF to Mini IDE - ~$10
http://cgi.ebay.com/Mini-IDE-CF-compact-card-adapt or-converter-for-notebook_W0QQitemZ8738065364QQihZ 005QQcategoryZ31534QQssPageNameZWD1VQQrdZ1QQcmdZVi ewItem

2GB CF - ~$30
http://search.ebay.com/2gb-cf-card_W0QQcatrefZC6QQ coactionZcompareQQcoentrypageZsearchQQcopagenumZ1Q QfromZR10QQfsooZ1QQfsopZ1QQftrtZ1QQftrvZ1QQsacatZQ 2d1QQsaprchiZQQsaprcloZQQssPageNameZWLRS

Google (write, spreadsheet, mail, etc...)

$250 - capable machine for everything but gaming.

Re:Flashy Mobiles

[Doc Ruby]

Now that is just what the Doctor ordered!

Since it was so easy to ask & get, I'm upping the ante, with a better order ;). The board looks like a direct mapping of traces from the IDE to the CF connector, with a few caps and a transistor. How come the many connector/formfactor flavors of Flash don't include native IDE? Then I could just plug multiple cards right into my machine, without $32 extra per card, without the extra space and weight of the PCB adapter?

And for dessert, how about a fanless PIII/1GHz/SVGA or better notebook without a HD, into which to plug those puppies?

Re:Flashy Mobiles

[Doc Ruby]

Those look good, especially the cheap PCMCIA/CF adapters. With a $175 8GB card, I'm probably done.

I think there is a demand for the benefits. But I also think notebook dealers don't market them (educate the market) because margins are still higher on 2.5" HDs, especially the ones bundled with new notebooks. Just unbundling those HDs opens competition from other HD vendors. And without market education, the unfamiliar products will find only niche markets, which also decreases dealers' economy of monolithic scale.

Then there's the recent cheapness of the new high-density CF technology. Dealers are squeezing the most they still can out of the already-amortized investments in small 2.5" HDs. They'll taper that pipeline as they increase the CF stream.

Sooner than later CF will offer the same superiority to dealers as to consumers. Especially the power/heat benefits will allow dealers to keep using cheaper, older CPUs for a few more cycles.

When all these cycles line up, in 5 years we'll have the equivalent of 2Kg P4/10GHz/128GB-CF running 24h on a fuelcell. Just in time to be obsoleted by a mobile "phone" making it look like a "portable mainframe" :).

Re:Flashy Mobiles

[Anonymous Coward]

Write speed to flash is way slower than write to IDE. And it gets even slower when your JFFS2 has to erase flash to free space before write. This is not big deal for mp3 players and similar applications because you do not write there too often but laptop and desktop usage is different.

I don't have solid numbers for you at this moment but I work with an embedded Linux box that has NOR flash for storage and it behaves all right overall but copying big files on it is a drag. Copying files from it is fine.

Re:Flashy Mobiles

[Doc Ruby]

Another poster posted $173 8GB "100x" flash. I assume that's 15Mbps, which is perfectly fine for the kind of mobile use I described. And 100K rewrites before failure is also plenty, considering it's removable media that will be upgraded for higher capacity as density:$ increases past that low maximum.

For a "desktop" (or any stationary storage), the volume, weight, power and heat issues are much less important. Which is a perfect complement, with bigger, cheaper HDs, on a network to a Flash mobile.

Re:Flashy Mobiles

[Anonymous Coward]

Here's another 2.5" adapter@ $18. Most of the pcb on these adapters is to make them mount hole compatible with a standard 2.5" laptop hdd. I've seen and own smaller adapters, but if I were to mount them in an hdd slot/laptop, they'd rattle around in the dead space.

To my knowledge, CF is IDE:
"That is, it appears to the host device as if it were a hard disk of some defined size and has a tiny IDE controller onboard the CF device itself. " -from wiki of course. Since you're interested, know that CF cards have a limited write cycle. Unless something has changed recently using them as a r/w filsystem ensures they die earlier than they were intended.

A quick google tells me the Sharp Actius MM10 was a fanless crusoe based laptop. Here's one on eBay, currently $182 If memory serves, a 1ghz crusoe is akin to a 700-850mhz p3. The Crusoe was a sippin' chip, and that lcd is fairly small. Combined with a solid state disk I bet the battery life would be great. Also consider used fujitsu lifebooks and toshiba librettos.

-e

Re:Flashy Mobiles

[Doc Ruby]

You're spoiling me :). You just made up for hundreds of abusive posts and moderations I've endured on this timesink website. Maybe that should have been an open-parenthesis.

Do you have any ideas about more appropriate filesystems for Flash drives than, say, ext{2,3}? The 8GB/100x Flash you recommended promises 100K write cycles before failure, but that's an average, so some will fail faster, requiring more writes to the remainder. Most Linux filesystems I know do a lot of rewriting to avoid fragmentation. Maybe a more indirect (extra pointer layer) allocation database that ignores actual fragmentation could consume the extra "seek time" gained from the more randomly accessible HW.

Re:Flashy Mobiles

[hotspotbloc]

The only problem is you could ruin the flash memory with constant writing. Now the fine folks over at Puppy Linux have worked out an interesting idea.

http://puppylinux.com/development/howpuppyworks.ht ml
http://puppylinux.com/flash-puppy.htm

Just curious, what model, ram and cpu is in it? If you go through with it start a JE and others can help.

Re:Flashy Mobiles

[Doc Ruby]

What I'd really like to see in bootable Flash drives are hybrids. A Write Once Readonly Memory partition for the installed OS, and rewritable Flash for the other working partitions, in a single USB part. Cheap enough (<$20) to discard (or just reuse the rewritable partition) when upgrading. Haven't seen any of those, at least none with the WORM partition writeable by USB.

Re:Flashy Mobiles

[Anonymous Coward]

Take a look at the $100 laptop web site. They're using flash instead of a hard disk and you'll find references to at least three different flast-friendly file systems under consideration. NAND vs. NOR flash matters too.

Re:Flashy Mobiles

[evilviper]

How come the many connector/formfactor flavors of Flash don't include native IDE?

A few certainly do. I have a cheap 44pin (laptop pin-out) 32MB Flash device I was using in my firewall for a long time. However, since CF to IDE adapters are so simple, and so cheap, it's not worth the cost of another form factor, especially since that limits your flexibility as well.

Bad idea

[JustNiz]

Does the drive automatically know and manage which files to put into flash(i.e. like a smart cache), or is this down to the OS to explicitly add/delete files in the flash?

If its the drive, then that sucks because the drive would need to know about the filesystems in use, and chances are it would only support Microsoft filesystems.

If its the OS that manages which files to put there, then it still sucks, as the drive and flash are combined.
It would be much better to have the flash as a separate component. Apart from the obvious benefit of being able to have it on a faster bus (such as PCI-E), flash memory is limited to about 100k rewrite ops, so when the flash is dead you have to throw away/replace an otherwise perfectly working hard drive too.

I guess if the only time the flash is written to is to update the boot files (e.g. as a result of an occasional OS reinstall or patch) 100k rewrite ops is not much of a limitation. However we all know Microsoft can't avoid filling up every little space with bloat files, so if it gets written to a few times every windows session the life of your hard drive will be pretty short.

Re:Bad idea

[pla]

Does the drive automatically know and manage which files to put into flash(i.e. like a smart cache), or is this down to the OS to explicitly add/delete files in the flash?

From the links (which don't give much detail), it sounds like the drive "looks" like a normal HDD to the machine. It just happens to have a nonvolatile cache built in, which means it can start serving files even before it spins up (which most likely explains the faster booting and restoring from hibernation); and for certain types of annoying "whack the disc twice per second" activity that Windows seems so very fond of doing, the drive doesn't need to spin up at all (thus the huge power savings).



If its the drive, then that sucks because the drive would need to know about the filesystems in use, and chances are it would only support Microsoft filesystems.

Not necessarily, for the same reason having 16MB of RAM cache on modern drives gives a pretty big performance boost - The drive itself just caches sectors, rather than whole files. Not only does it not need to "speak" NTFS that way, but if you have a 100MB file of which only a handful of sectors get used 99% of the time, it can just keep track of what actually gets used. With a RAM cache, the big boost comes from not needing to seek as often; with these hybrid drives, you get that same perk along with not needing to spin up the drive so often.

Re:Bad idea

[zidohl]

Hopefully the OS would have some controll over this enitiy apart from the HD itself, so you could disable the use of it if nessecary.. This would help avoid the shorter lifetime of such a harddrive, so if the flash-dive fails it would go back to be a "normal" harddrive. If it was controlled by the OS it could be a possibilty to say, tell it to only increase boot performance, which would increase the lifetime of the flash-drive by alot.

Re:Bad idea

[JustNiz]

So if this is just a big cache then the no. of write ops. to flash is a real problem. Also, how would the drive know which sectors are needed at boot time in order to pre-cache them?

Re:Bad idea

[Superfarstucker]

Modern flash devices sport a million write/read cycle and defect levelling, the controller keeps track of how many times each sector has been written to and compensates accordingly, since files do not need to be contiguous on a flash drive. In any event, your better off replacing your hard drive at least every 5 years, more like 3, so it won't bother me if it wears out this fast.

Re:Mbytes?

[OneoFamillion]

> Are those 6 minute MegaAbs?

No, not six minutes, SEVEN! No one could get a good workout in just six minutes, duh!

Hmm... MegaBS = 1000^2 people bullshitted? Could come in handy with all these RIAA topics.

Flash vs traditional materials

[porkface]

Does modern flash memory degrade more quickly per write than hard drives?

How much faster is flash storage memory than hard drives?

While 256MB would still speed boots for hibernate files larger than 256MB, current boot speed of a 256MB hibernate file from a hard drive is nearly instantaneous anyway, negating any real value to this. The real value would only kick in for systems with more than a gig of memory.

Re:Flash vs traditional materials

[swv3752]

The point is that for the first couple seconds it reads off the Flash, and lets the hard have a slow spinup instead of the high power draw instantaneous spinup.

Probably a good while yet

[Sycraft-fu]

Flash is getting better at an amazing rate but it's got a looooong way to go to catch HDs. You need more capacity, much less cost, and also higher speeds. While flash has faster random access, it can't hit the sustained transfer rates of HDs, at least not the normal flash RAM you find for sale everywhere.

I imagine the hybrid HDs will be the first step. Try and get the best of both worlds. A small flash store for frequently accessed thigns to get lightning fast random access, a large magnetic disk so you don't compramise on storage. Windows Vista is apparantly going to be pushing this rather hard. MS notes support for it as one of the features, and even if you lack a hybrid HD, you can get something similar by giving it a USB flash drive and instructiong Vista to use it as an app cache. Parts of programs are then put on the flash to speed load times.

I think that's the kind of thing we'l see for a number of years here until flash gets cheaper.

Re:Probably a good while yet

[MikeBabcock]

Personally I think hybrid HDDs are a bad idea. I plan on keeping a new 300GB hard drive for quite a while, even after upgrading my RAM, motherboard and CPU. That said, if I can get a huge performance increase by boosting the flash space on the hard drive, why not just add CompactFlash drives to the system instead? Use semi-permanently attached USB keys for swap space instead of the hard drive for all it matters; you can get a gig for about $25 these days. That's a much better deal than buying a new hard drive to take advantage of this hybrid technology.

Re:Probably a good while yet

[Anonymous Coward]

Just apply striping over several chips and sustained speed problem is solved.

J.K.

Re:Probably a good while yet

[owlstead]

I see this as a step up to a full operating system, including runtime code for applications, fitting on a flash drive. HDD's would then be used for media storage and backups. Almost all of my data consists of media and/or backups. Say, a 16 GB flash drive would be enough to remove my hdd for every minute use. Then it would just need to spin up when it's needed, saving huge amounts of power.

drives are faster, too

[SuperBanana]

Hibernation works by writing the contents of the RAM to the hard drive, so this would only work if you had = 256 MB RAM. I don't think too many new systems meet that requirement, and even less will after Vista comes out

Not to mention your average notebook hard drive these days is fully capable of pushing 20+MB/sec for the linear read a "resume" requires, unless the hibernation file is fragmented. Even fairly expensive media like Sandisk Compact Flash "Extreme III" cards for digital cameras can't hit that, and one of those (1GB) costs about the same as a 100GB hard drive. Silly.

My Macbook by default hibernates, but I found a setting to flip that off so that it "sleeps" like it should (involve the 'defaults' command, I forget exactly.) Now it takes about 2 seconds to 'wake up'. Ironically enough, hibernation takes longer than it takes to boot (about 25-30 seconds) and the scale has probably been tipped even further in favor of "booting" with another GB of ram I just added; by my rough calculation it'd take well over a minute if most memory was in use at time of hybernation (maybe the OS clears out all disk cache before doing it- you'd hope so.)

Hibernation is for when your battery is pretty much dead and the laptop wakes up to hibernate before it looses the contents of RAM due to battery failure...and can people REALLY not wait the time it takes to boot or wake up from hibernation and copy the data back into RAM? Yeesh.

This seems like an attempt to save themselves in a market they're just not competitive in. From all accounts I've seen (and personal experience), Seagate's ATA-drive reliability is in the trashcan these days; the 7200.8 was a fiasco, and the 7200.9 doesn't seem much better. IBM sold off their drive business (which was a market leader in almost all segments) after the Deskstar/Deathstar fiasco, but Hitachi seems to be doing fabulously. I had a 7200.9 300GB drive that died within 12 hours of operation. It's been RMA'd, and the replacement will be sold on Craigslist or similar. In the meantime, a shiny new, cheaper, cooler-running, quieter Samsung Spinpoint is sitting in its place.

I think Seagate has seen the writing on the wall- hence the merger with Maxtor. I would imagine you'll see them merge Seagate/Maxtor technology in their ATA line and sell exclusively under Maxtor, and Seagate will go back to being a mostly SCSI brand, as their reputation there seems intact.

Re:drives are faster, too

[NeMon'ess]

Congrats, your post gives the impression you're basing your opinion of Seagate quality just on a bad experience with one 7200.9 drive. How very scientific and valid your statistics are. StorageReview.com is much more comprehensive.

Re:drives are faster, too

[Anonymous Coward]

It sounds like OS X hibernation is about where Windows 2000 hibernation was. It's surprisingly faster in XP.

The fastest camera flash memory (Sandisk Extreme) can sustain reads of about 12MB/s, and it's possible that Seagate's may not reach these speeds, but don't forget that your average HD reads at 0MB/s while spinning up and seeking. The ideal thing to do is put all of the data that must be read first (while the disk is spinning up) and all of the small reads (where throughput is dominated by seek time and rotational latency) on the flash memory. Then the disk can supply all of the linear data while the flash supplies everything that would slow the disk down.

The biggest reason for adding flash isn't to speed up booting, it's to prevent the disk from having to spin up during normal use. In theory you could accumulate 256MB of writes before having to spin up the disk. You might be tempted to suggest just caching those writes in memory, but maintaining the integrity of the filesystem requires that some writes be committed before others are allowed to proceed. Having the flash memory will greatly increase the performance of journaled filesystems while reducing power needs.

As for price, 256MB is 2Gbit, which shouldn't cost Seagate more than $5 per chip in the quantities they're buying.

dom

Re:drives are faster, too

[Stellian]

your average notebook hard drive these days is fully capable of pushing 20+MB/sec for the linear read a "resume" requires, unless the hibernation file is fragmented. Even fairly expensive media like Sandisk Compact Flash "Extreme III" cards for digital cameras can't hit that

Yes, but how much of that is due to the serial interface?
So let's assume a flash array can be written at 10MB/s. You can partition you flash chip in multiple arrays and write multiple blocks in parallel. You can use multiple flash chips in the drive.
So, with a few simple tricks the speed of flash can increase by one order of magnitude, limited only by what ATA/SATA can provide (100 - 150 MB/s). Plus, no fragmentation.

Re:drives are faster, too

[loraksus]

Although I like the 5 year warranty and will probably buy their drives in the future exclusively because of that (I have a fair number of drives, replacement costs do add up) I did have to wait a little bit over a month and a half for them to get a replacement 300 gig drive in stock and they still didn't have the drive available when I called, complained and sent me a 300 gig 10k scsi drive. Hard to be pissed about that, but to me it looks like they have had beaucoup rmas on the 7200.9 drives.

Re:drives are faster, too

[networkBoy]

You still get fragmentation (the addressing is not progressive and has gaps), but you no longer care because it costs you no speed.
-nB

Re:drives are faster, too

[Anonymous Coward]

Seagate's ATA-drive reliability is in the trashcan these days

Oh, is that why Seagate is the ONLY ONE to offer a five year warranty? Not even your Crapbook Pro offers that. Of course I should expect such nonsense from an Apple fanboy. I have a suggestion fruitboy, why don't you stick a banana up your ass?

Re:drives are faster, too

[Anonymous Coward]

Mac OS X is actually far batter than XP when it comes to hibernation - especially since it doesn't treat hibernation and sleep as separate things. It's hibernation is sleep 2.0, if you will. (Well, Apple calls it Safe Sleep.)

The reason I say it's better is because the OS only re-reads RAM from disk when it has to. When you close the unit or choose the "sleep" command, memory is written to disk. When you wake from sleep, the computer intelligently does a normal wake-up or wakes from hibernation depending on the state of RAM.

Also, regarding the time question, I have a PowerBook G4 DLSD, which was the first model to officially support hibernation. Before composing this reply, I put it to sleep, unplugged it, and took out the battery for a couple minutes, so the backup RAM power would be drained. (Yes, there's backup power in case you need to swap batteries while on the road. It makes battery swaps very fast and painless.) Upon plugging the unit back in to power and opening it up, the unit came to life. In three seconds, my screen came up with a grayed-out preview of the running OS. In a total of 25 seconds, the computer was back up and responsive.

This was times with my phone's stopwatch, with 1.5 GB of RAM and the stock 5400 RPM 80 GB hard drive. After bringing the machine back up, Activity Monitor reported over 1 GB of RAM in use after bringing the machine back up - 167 MB wired, 487 MB active, and 474 MB inactive. Only 394 MB were free, and total VM size was 12.31 GB. In other words, the machine was in use, and had been for a while. Not the most favorable conditions for a fast wake, but more realistic than many other tests might be.

Windows ends up being slower to wake from hibernate simply because it takes the computer 10-15 seconds to get through the POST. In fact, let me compare for you...

I just ran a similar (as similar as possible) test with my WinXP SP2 desktop. AthlonXP 1800, 768 MB, 18.2 GB primary drive, an Atlas 10K III. In other words, a very favorable comparison - machine with faster system bus, comparable CPU, half the RAM, and hard drive that is twice as fast at the very least. Look up the specs on the drive, it was a screamer a few years ago and is still pretty quick. Windows took no less than a full minute to resume. 21 seconds from power-button to main-BIOS-POST finish, 13 seconds for the SCSI bus scan, and 26 seconds for the remainder of the process to the login screen. Even if you discount the first half of the process, as it's outside MS's control, the part that Windows was in control of was notably slower. This is with a machine that outclasses my PowerBook in pretty much any way that should matter for this test.

Seperate battery backing for RAM?

[swb]

I guess I don't see why they're using flash. You'd think that they could accomplish it with some RAM module with its own seperate battery backing (like a RAID card), and then have the disk writes delayed to some optimum amount to minimize power consumption as well as being used as a pre-fetched RAM cache with the same optimization, with the read/write split dynamically reallocated as needed.

Re:Seperate battery backing for RAM?

[Anonymous Coward]

So you're wondering why they use a cheap and reliable technology like flash instead of an expensive and unreliable technology like battery-backed RAM? Probably for the same reason that you can get a 1GB flash memory stick that sits on your keychain but you can't get a batter-backed RAM chip that sits anywhere.

Surely you've noticed that hard drives come with no more than 16MB or cache, while this new drive is going to come with 256MB of flash. Using battery-backed RAM is an expensive, special-purpose application. You would only use it where the expense is justified and the battery is only for power backup (like in the case of a RAID controller where power may be out for a few minutes or hours).

LSI lists battery specs at 72 hours for a 256MB cache and suggests replacing the battery every year. This battery is both bigger and more expensive than a typical laptop hard drive.

I hope this post answers your questions.

dom

The best hybrid option out there

[RyanXP]

I'm really looking forward to trying this out in a notebook... my cousin Phil tried putting a hybrid in his mini tower, but the whole plan backfired and he ended up horribly disfigured. Apparently Grolar Bears are SATA, and when he tried plugging it into his older moboard it got pissed.

Hydrogen?

[edmicman]

When can I get a hydrogen fuel cell hard drive? Or does it use regernative kinetic energy from the platters spinning to generate power?

How long...?

[Anonymous Coward]

Very long.

Flash lifetime?

[JensR]

Using flash chips as harddrive replacement comes up now and then, even if in this case it would be just a cache. But what about flash lifetime? Last time I checked a flash chip could only be erased something like 200.000 times, which could be used up quickly in normal operation. Or would the flash area show up separately?
In that case it may be easier to get one of those IDE/compact flash adaptors and have the flash as a separate device.

Re:Flash lifetime?

[vidarh]

Keep in mind that the stated erase cycles are per erase unit, and each erase unit is rarely more than 128KB. You can write multiple times to each erase unit too, so you can write a block at a time to each erase unit, and as long as you run a sensible filesystem or a flash driver that remap blocks properly the number or erases will be fairly low and can be reasonably evenly distributed over all the erase units.

Re:Flash lifetime?

[JensR]

Other people mentioned the defect management that is part of a normal HD firmware, that I didn't think off...

Mod this up, please

[Weaselmancer]

These are good questions.

Want some cheese?

[Inoshiro]

Worry away, but I don't think you're going to lose any precious data any time soon. I can give you two reasons why:

1) "When compared to a hard disk drive, a further limitation is the fact that flash memory has a finite number of erase-write cycles (most commercially available flash products are guaranteed to withstand 1 million programming cycles)" (Flash memory limitations).

2) 4 years of 24/7 operations is 35,040 hours of use. That's about 28.5 writes/hour, or a write every other minute for 4 years. Chances are you'll upgrade before 4 years.

And, let's face it, people making HDs aren't stupid. If they detect NAND failure, they will write through to the HD itself and disable that part :p

Re:Want some cheese?

[Chirs]

Okay, I sat down and worked out a worst-case scenario.

It's got 256MB of flash.

A reasonable write speed for a drive is about 30MB/sec, so we write to the whole flash in 75 seconds. Assuming 1 million writes per sector and perfect wear-levelling, then the flash will be dead after about 2.4 years of continuous use.

So, a busy server might kill the flash in this drive before the warrenty expires. For most people it probably won't be an issue.

How to do this with Linux

[WhiteWolf666]

1. Acquire Flash memory. USB or whatever, it doesn't matter.
2. Insure you have the correct interface connections to the computer (USB port, USB cable, CF/SD drive, weird built-in hybrid device).
3. Boot Linux
4. Find location of Flash device. A modern distro will point this out to you on the desktop.
5. Use your GUI partitioner to define the flash device as your swap space. Be sure you purchased a flash device with size > system ram.
6. Suspend2Disk really, really fast.

Also, given a reasonably long up-time, enjoy the perks of a system with high-speed swap space. Applications, data, kernel; whatever! It all gets faster! Be sure to crank up your swappiness value for maximum effect; this'll have Linux swapping out just about everything it can get its hands on.

Given a modern flash device, with 1 million or so read/write cycles, and defect balancing, even under very high-usage you should get years of use.

Re:How to do this with Linux

[mailman-zero]

I would love to see benchmarks using a USB flash drive as a swap partition. Say, 2GB flash drive with 1GB RAM. Why didn't I ever think of trying this?

Note to self: Buy big USB key.

Re:How to do this with Linux

[tknd]

I actually did this with my linux box. I'm currently only storing the log files on the flash drive. I bought a few CF to IDE converters that make the compact flash card look like a regular disk to the computer, rather than a usb device which is actually pretty nice. (In another experiment I was able to get DSL to boot off of a CF card with a DSL hard drive installation to the flash disk and bam, I had a computer without a real hard drive.) I did this in order to reduce the amount of access to the hard disk so the hard disk could remain off more often with a little bit of power savings and hopefully a little less wear on the disk itself. The raw throughput of my CF card is a little slow because it was a cheaper 1gb card.

Ideally I'd like to get a CF card to have enough capacity to run all of the software and OS specific files while the hard disk just acts as a backup and file storage device. It's quite nice not having to listen to the hard drive hums and clicks while reading data and makes a silent computer possible.

Re:How to do this with Linux

[aunitt]

Err.... surely it's a lot easier, cheaper and faster to just buy more RAM then you don't have to swap?

Or is this supposed to be a "funny" post and the moderators are asleep?

Re:How to do this with Linux

[WhiteWolf666]

A little bit of both.

It's a stupid idea, but then IMHO, the Microsoft implementation is a stupid idea, too.

The big deal is that Microsoft is pushing non-volatile RAM as an alternative to disk-based swap and disk-based hibernation. Given that Linux "sees" disks as disks, it doesn't matter whether you use platters, flash-ram, or packet-based DVD-ROM as swap space.

If you've got spare flash ram lying around, its not totally stupid ;-)

Re:How to do this with Linux

[evilviper]

I did this in order to reduce the amount of access to the hard disk so the hard disk could remain off more often with a little bit of power savings and hopefully a little less wear on the disk itself.

Having to spin up/down often will result in far more wear-and-tear on your hard drive, and shorten the lifespan.

It's quite nice not having to listen to the hard drive hums and clicks while reading data and makes a silent computer possible.

Any reasonably new hard drive has a very quiet fluid bearing motor, practically eliminating the hard drive whine, and dramatically reducing noise. As for the clicks, you just need to use hdparm to set the accoustic noise management to the quietest possible setting, and you won't hear it ever again.

Though my DVR has (slow) fans and a hard drive, it's still damn-near silent. I just wish I could find a silent, slot-loading, DVD-ROM, though the Pioneer drive is acceptable if you force it to stay at 3X with the Windows software.

Benefit for linux?

[Japi]

I don't see this device being all that useful for hibernate wake-up, but it is interesting if the 256MB just acts as a persistent write-cache. Often in linux the hard drive never spins down, because things are constantly being written (log files, atime updates on the FS, etc). This can be reduced by using laptop-mode (and mounting -o noatime ) for the filesystem, but then you run the risk of losing data if the system dies and nothing has been synced with disk for a long time. And the last thing you want to do is spin up the disk every 10-15 mintues(because of the wait, power, and wear from this), but you need to to keep the filesystem consistent. But with 256MB of write space, that doesn't require the disk to be spun up, that is synced with the "real" disk when it *is* spun up (for reading uncached stuff probably) would make for a large power savings on systems that have enough RAM to keep everything in memory. I guess linux can take advantage of that immediately, with no special tweaks, if the drive is doing everything. But I have read other articles that say that you need Vista to take advantage of this drive.

Re:Benefit for linux?

[Adelec Bakkal]

"as a persistent write-cache" wouldn't this just kill the flash? if we consider the limited WRITES... I'll try to find some more precise specs of this hybrid drive, just to know how long it can withstand the write processes. If anybody do find some, please post!

OT: Hybrid Tax credit not deduction

[Anonymous Coward]

Don't be confused about the 2006 US tax benefit for buying a hybrid vehicle. It's a tax credit, not a tax deduction, which means it's subtracted from the tax you owe the government, not the amount of income you made. It's tax(income())-credit instead of tax(income()-deduction).

http://www.cnn.com/2005/AUTOS/tipsandadvice/12/21/ hybrid_tax_credits/index.html

Where does it say it exceeded esitimates

[lapagecp]

"Interestingly, the new Momentus 5400 PSD has also exceeded earlier estimates of hybrid hard-drive performance, which said that such drives would add an extra hour to the typical battery life of a notebook PC." Read carefully people. The article says that seagate excpected a 9% reduction in overall power use. They then said that what was observed was a 50% reduction in harddrive power use. Unless they specifically compare hardrive power use to overall power use all we can infer from this is that a 50% reduction in harddrive power use reduces the total use of the system by 9%. Lets not forget the bright shiny candy like screen and whirly DVD rom and flashing lights and sounds that laptops make.

Re:Where does it say it exceeded esitimates

[Xichekolas]

My laptop doesn't have flashing lights...

... I got screwed!

Regression??

[Anonymous Coward]

Isn't this a step backwards? Gas and electricity powered hdd's?

*duck*

Hybrid you say?

[jscheelmtsu]

Yeah, but how many megs-per-gallon does it get? If it's a hybrid, it should get at least 60 in the city. Of course, you folks that live out in the boondocks won't get nearly the same efficiency.

Rewrites and other musings...

[Xichekolas]

First, the rewrite limit on flash nowdays is in the low millions... so even if you could only rewrite a million times, and you did it 100 times a day (about once every minute and a half)... the drive would last over 27 years... Not sure why everyone keeps bringing up this longevity issue. Sure, don't put temp files or virtual memory on it... every time Windows pulled a thrash maneuver you'd lose 8% of your HD lifespan ;) ... but for storing OS boot files and commonly used system files, the lifespan thing isn't a problem... That said, I think they should put a flash socket on the mobo so the stuff is upgradable if you want larger capacity. That way, you wouldn't have to buy a new HD to get more flash memory or vice versa...

Re:Rewrites and other musings...

[mrvan]

Uhm... My day has 24 hours x 60 = 1440 minutes, making 100 times per day about once every 15 minutes. (unsurprisingly, since 15x24 is close to hundred) So if you really need to write every minute, the lifespan would be about 2 years, which is serious (1E6/(1440*365) = 1.9) If it is really used as a cache as swap, I'm guessing it'll be closer to once per second, making the lifespan a couple weeks(a million seconds is less than two weeks), so you're quite right about the virtual memory and temp files

Re:Rewrites and other musings...

[Xichekolas]

Yeah... uh... somehow I took 86400 (secs in a day) and divided by 100 and got 86.4 instead of 864... so uh... thanks for calling out my stupidity...

Mods, just mod me down into oblivion...

*goes and sits in time out*

Re:Rewrites and other musings...

[robertjw]

(unsurprisingly, since 15x24 is close to hundred)

Huh? Did you mean 4x24?

Re:Rewrites and other musings...

[vidarh]

The number of erase cycles PER erase unit is typically in the millions. Typical flash units have erase units of anywhere from around 256 bytes to 128KB. Assuming 256KB, to make it even wrose, that means 1K erase units in this unit. Assuming each of them can be erased one million times, erasing a 256KB block per second still means a lifetime of more than 3 years.

And keep in mind that is erases not writes. A typical flash based system will remap blocks so that no erase units are erased until the unit is more or less full (and you can write multiple times to the same erase unit - you just can't reset any bits you have flipped without erasing the whole erase unit).

That's not how hibernation works.

[ivan256]

Hibernation works by swapping out every possible page, then the remaining memory contents is written to the hard drive. You don't need your RAM to be 256MB or smaller to see the performance boost, you only need the system pages to be that small, because you can be fully awake from hibernation without swapping those pages back in. That takes us to the other thing you said: most modern operating systems *do* fit in 256MB. At least, their text segments and essential data segments do. 256MB is actually *mammoth* in that context. In most cases, you should be able to contain the kernel and essential libraries in 256MB of flash, and still have room left over to write your system pages to during hibernation.

failure rates

[Ender Ryan]

All this new harddrive tech is cool and all, but what I want more than anything else, including more and more massive storage, is better reliability.

In the past year or so, I've had, literally, 50% of all drives I have purchased fail. Mostly Western Digitals, FWIW.

For that matter, I've had several expensive raid controllers fail too. This shit is really starting to piss me off.

Re:failure rates

[WuphonsReach]

All this new harddrive tech is cool and all, but what I want more than anything else, including more and more massive storage, is better reliability.

In the past year or so, I've had, literally, 50% of all drives I have purchased fail. Mostly Western Digitals, FWIW.

For that matter, I've had several expensive raid controllers fail too. This shit is really starting to piss me off.

Odd, because I haven't seen a drive fail in quite a while personally (and I probably have 20+ spindles sitting next to me in my office). I have a mix of Seagate, Maxtor, Hitachi and Western Digital.

As always, the #1 killer of drives is heat followed by poor power. Mechanical damage is also possible, but not likley in stationary systems.

So, what operating temps were you running at? Were you using a line-conditioning UPS to correct over/under voltage situations? Was your power supply a no-name cheap one or a quality one with good tolerances?

Re:failure rates

[Gat0r30y]

Seagate drives come with a 5 year warranty. WD drives do not.

Huh?

[Jeff DeMaagd]

At a max of about 2.5W on write, a notebook hard drive isn't the biggest power draw in a notebook. Idle power is maybe half that.

You have the screen (flourescent backlight) (likely tens of watts) and the CPU (Intel Core Duo is 31W), probably the GPU too. Cutting the CPU to an LV chip (Core Duo LV is 15W) might give you a two or four more hours, depending on the display and the GPU. Don't tell me that saving one watt is going to save an hour of power on battery time.

Re:Huh?

[cnettel]

Ehm, an Intel Core Duo only draws 31 W when you are at 100 % CPU usage. As a comparison, my complete Dothan system uses around 15 W when idle, while the CPU alone is specced to around 25 W. I can bring it up to 40 W by turning backlight all the way up and start some CPU intensive tasks. (I've not bothered measuring while playing games from battery with the DVD going, but I know for sure that this would give even higher numbers.) If I really turn the backlight down and forces the CPU to go at 600 MHz, closing all processes, I get 12 W. No, I wouldn't save an hour by turning the HD off, but I would at least save something, and your numbers are grossly inflated, as you assume 100 % CPU load.

Re:Huh?

[swillden]

Don't tell me that saving one watt is going to save an hour of power on battery time.

Running with the drive spun down saves over an hour of power on battery time on my Thinkpad's extended life battery. ~6 hours with the drive spun up, ~7.5 hours with the drive spun down most of the time. Both figures assume the CPU running at low speed (600Mhz in my case), the backlight dimmed all the way down, and relatively low CPU load. Well, that's with a new battery. Mine is a couple of years old now and doesn't do anywhere near that well.

the best part...

[Mister Whirly]

And the best part of the new hybrid drives is that you will grow to love the smell of your own farts...

http://www.hybridcars.com/south-park-hybrid-smug.h tml

Correction

[EmbeddedJanitor]

Writes are in the hundred thousand range. To get these densities/price points, this will be NAND flash.

You would not be wanting to use this space for anything like swap space because NAND write time is so slow (compared with platter). It would probably be good for storing/buffering multimedia files so that you can shut the disk down while you're playing.

Re:Regarding the other announcement, DB35 series..

[tppublic]

I'm guessing the drive security tools quote refers to the same security technology in the Momentus 5400 FDE:

Seagate's solution is hardware based, meaning the encryption functions are performed on the drive, separate from the operating system. Even the password or user ID is encrypted and stored in an inaccessible area on the drive, providing much stronger security than today's software-based encryption solutions.

A system with Momentus FDE comes fully enabled for encrypting all the data on the drive, so there are no time-consuming installation and configuration requirements as with software; users simply enter their ID, and if authenticated, they have full access to their data. Since the user ID is entered before the operating system can load, it's essentially impossible for any spyware-like code to have visibility of the key.

So I would say for a DVR manufacturer, this helps prevent things from being recorded and then the drive pulled to steal the video.

It's been done...

[SanityInAnarchy]

Suspend2 for Linux will do lzo compression before writing to disk. Also, I believe they're moving towards being able to store/resume an image to/from anywhere, and have the resume operation be triggered in userspace (from an initrd).

Also, with a fair amount of memory on a laptop and a good filesystem (or Laptop Mode on Linux), you don't need this Flash device to avoid using the disk. Problem is, I've never really gotten it to cache much of the music, although it will avoid writing until it has to, even if I "save" -- which is fine, because the OS can be pretty stable, and a laptop has built-in battery backup, and I can always run "sync". Now, if only I had Linux running on my Powerbook... see, HFS+ does write to disk as soon as it can, which is good for saving data, but bad for saving battery.

Re:It's been done...

[swillden]

Problem is, I've never really gotten it to cache much of the music, although it will avoid writing until it has to

Given a relatively small music collection, and a relatively large amount of RAM, you can just do something like:

while true; do find ~/music -type f -print0 | xargs -0 cat > /dev/null; sleep 60; done

I've done something similar while watching movies on airplanes with laptop mode turned on, and when my Thinkpad T40's extended life battery was new I could get over seven hours of movie-watching out of it. That's very useful on a flight from San Francisco to Sydney.

What would be ideal would be a music/movie player that understood laptop mode, and pulled as much of its playlist into RAM as possible. I toyed a bit once with modifying xine so it would pre-read a gig or so from a DVD into RAM while playing it, so the DVD drive could spin down. I figured since my DVD-ROM can read the data in at 8x, and since few (if any) movies use the full 1X DVD bandwidth, the DVD-ROM drive should only have to spin a small fraction of the time I'm watching the movie. I didn't put enough time into it to get it to work, though. The same idea would be even more effective at keeping a hard drive spun down.

Re:It's been done...

[paropaco]

Now, if only I had Linux running on my Powerbook... see, HFS+ does write to disk as soon as it can, which is good for saving data, but bad for saving battery.

Did you try "sudo killall update" on Mac Os X? It works for me when I'm on the go, but then I'm only reading data, not writing (it's only these darned file access times that make it necessary to kill update in my case).

Replacing your primary hard drive

[KenAndCorey]

What is the data-transfer speed difference between a flash drive and a regular hard drive? For instance, you could buy an ATA flash drive and install your O/S to it, and maybe your applications. Then have a normal drive for your temp files, data files, etc. Would the O/S boot faster from the flash drive?

Re:Mbytes?

[gavriel407]

You must work for one of the hard disk companies; we like our "Mega" to mean 2^20.

Here's the math...

[codemaster2b]

These were their working definitions:
- 4 games/8GB or 2GB/game
- 8hrs video/8GB or 1GB/hr video
- 133 hrs music/8GB or 60MB/hr or 128kbit
- 2560 photos/8GB or 3.2MB/photo


Thus here is the math: - 750GB HDD - 300 GB left over
- 450 GB HDD = 15000 songs + 1500 photo + 50hrs video + 50 games + 25 DVDs
- 450 GB HDD = 60GB songs + 5GB photo + 50GB video + 100GB games + 25 DVDs
- 235 GB HDD = 25 DVDs
- 1 DVD = 9.4 GB

I guess they really mean it. Of course, the only way you're going to get a DVD onto your hard drive is through... um... antiquated software.

Re:Here's the math...

[Anonymous Coward]

Actualy I can tell you the average DVD is around 6GB --- and its real easy to rip them... uhh a friend told me ---if your too lazy to encode them that is. The average Blu-ray movie will be around 35-40 GB (i think), that means only 20ish movies on a 750 GB drive yikes.

Re:Here's the math...

[iainl]

The "Average" BluRay movie is only a single-layer disc, so 25Gb max. There are rumours floating around that they might have Black Hawk Down lined up for the end of the year as a test run of dual-layer discs, but so far they're unconfirmed.

So far, the HD-DVD releases have had video at around the 20Gb mark, and then a selection of different audio formats. Every company going dual format is using the same video encode for both, and then playing with the amount of audio and extras to make them fit appropriately.

Re:Here's the math...

[codemaster2b]

There are two DVD formats, DVD-5, and DVD-9. DVD-9 is dual-layer. Thats 9 as in 8.5GB. And it IS real easy to rip them. I am fully capable of doing so (both encoding, and pure copy). It is, however, illegal to do so for commercial DVDs. Specifically, it is illegal to do so in America where the MPAA writes the laws. However, Samsung in Korea may not have the same restrictions. I am not in a position to know.

Samsung hybrid drive sector pinning

[rolofft]

What I've read about Samsung's variation of this technology is that it is accessed somewhat independently, at least by the OS. The OS determines sectors that deserve quicker read access and "pins" them to the flash drive.

Flash card slot on drive?

[John Bokma]

I would love to see this to become part of desktop drives as well. Moreover, it would be really nice to have a drive with a slot one can insert his/her own card into. Especially if the flash memory is significantly faster compared to the harddisk. Almost we are 20 years back: Acorn Archimedes, OS in memory :-D. Power on, beeeep, desktop.

Re:Flash card slot on drive?

[Source Quench]

Acorn Archimedes ahhhh. Beeeep! "out of memory" grrr.

Re:OT: Hybrid Tax credit not deduction

[redalien]

For a deduction simply use the magic words "business expense", and tada!

Compact Flash 'wearout'

[david.emery]

This topic came up on my program recently. The figure we were quoted as "best available practice" is 100k writes. Pointers to higher performance specs would be much appreciated. (Since my program is a long-lived embedded system, it's clear that using flash memory will result in having to replace the 'flash drives' periodically, and that's taking the device out of the field and into the shop, or at least sending a tech with a replacement part.)

But consider: 100k write duty-cycle, over a 3 year period, is an update rate of about 90 seconds. That means it's probably OK for user data, but clearly not OK for swap or for system usage such as inode tables for the file system... At 1m duty cycle, that goes down to 9 seconds, which is getting into the ballpark for system kinds of writes (e.g. inode updates for the file system), but it's still not there for swap.

But the underlying problem I'm having is recovery from an error. My guess is that you have to 'write then read' to verify that you have NOT hit the error, and that the probability of the read failing is much less than the probability of the write failing. (And I believe that reading is much more reliable than writing, so that's probably av valid assumption.)

What you then need is a recovery strategy for a failed 'write location'. I guess you could use current failed sector techniques.

So I think this is a cool idea, but I still have some questions about the end-to-end performance and reliability.

        dave

MOD UP PARENT

[Anonymous Coward]

If the firmware on these drives is designed in a reasonable way, you will get this advantage right away. As it can (and should) be transparent to the OS, it will work with any OS. And yes, disks spinning up all the time to write small changes to the disk is a problem I often encounter with Linux. (I guess the problem exist with other OS as well, but I wouldn't know since I don't use them).

Re:Mbytes?

[multiOSfreak]

Are those 6 minute MegaAbs?

I think the OP meant megaBS.

It would be more useful

[phorm]

As a flash-module that snapped in beside a hard-drive, rather than with it. That would make it quite a bit easier to upgrade, etc.

On the other hand, the upside is that you wouldn't necessarily need a seperate IDE channel/cable for it this way.

Owned.

[DoctorDyna]

Personally, I'd rather wait and go with this:

http://www.ddrdrive.com/ddrdrive_prototype.html

Ditch the hard drive completely. Plug this into a UPS by itself and you've got the safest, fastest drive. Period. Imagine how long this would last if it was the only thing plugged into a 1000 va UPS? Spend the time you save doing a weekly backup to a couple of DVD's and you're done.

Re:Owned.

[vidarh]

There's lots of RAM based HD "replacements" available today. The problem with them is that they are ridiculously expensive, and have a very low storage density (even assuming 4GB or 8GB DIMM's, it'll take far more physical space than a harddisk). It's not worth it unless you have apps that are entirely limited by disk IO.

db usage

[Anonymous Coward]

ooooh. I wonder what the write latency on one of these is.

I have a low end database server that would just love having essentially a flash cache for the writeahead logs.

oh really?

[dirtyhippie]

Samsung forecast that the first hybrid drives would ... reduce power consumption by about 9 percent overall, increasing a notebook's battery life by about an hour.

Uh... Someone in Samsung's PR division does not realize that the typical laptop does not get 11 hours of battery life. There has got to be a way to hold PR folks accountable for the stupid and wrong things they say.

I'm still waiting on the "hybrid" PDA!

[kimgkimg]

When the heck are they going to apply this technology to the stupid PDA!?!?! I just love having to reload and reinstall every single !#%$@ thing ever time I happen to leave the thing in a drawer for a few weeks. Totally useless as a occasional reference device unless you're using and charging it every single !#@$@% day!

Re:I'm still waiting on the "hybrid" PDA!

[sunhou]

When the heck are they going to apply this technology to the stupid PDA!?!?! I just love having to reload and reinstall every single !#%$@ thing ever time I happen to leave the thing in a drawer for a few weeks. Totally useless as a occasional reference device unless you're using and charging it every single !#@$@% day!

I don't know if you'll see this since the thread is 12 days old, but I just got a Palm Tungsten E2 which uses flash memory, so if the battery runs down it doesn't lose its contents. That's quite nice, since with my previous PDA (Handspring Visor), I had the same worry as you.

256Mb...

[Source Quench]

...ought to be enough for anybody!

Could help NTFS?

[Werrismys]

When handling directory entries NTFS is so slow it's not even funny. Maybe with some extra glue the drive head trashing related to NTFS metadata handling overhead could be targeted -> flash ... -> profit. Or something. Try this: http://www.fs-driver.org/ Then try mirroring a big directory structure to/from NTFS ... and to/from ext2/ext3 using this driver. The performance difference is enlightening.

I don't get it.

[billcopc]

How is this better than just having 256mb of extra ram ? Since we're talking notebooks, there's a battery backup so having a big r/w disk cache is not a problem, your data won't vanish during a brownout (america's national pastime).

Hell, I have 4 gb ram in my desktop and it's the best damn speed booster I've ever had. The OS caches everything and syncs whenever the PC is idle, letting me run my bursty, data-intensive processes at full speed without any concern for disk throughput.

I think the real reason why I'm suspicious is the volatile nature of hard drives. Why would I want a hybrid drive that needs replacing every few years, when the part that breaks down is the cheaper half of the device ? Seems like a way for the HD manufacturers to squeeze more profit out of their mediocre products.

Re:Mbytes?

[Anonymous Coward]

Nah. Mega = 10^6, Mebi = 2^20. http://en.wikipedia.org/wiki/Binary_prefix

It's because vendors sell different things

[billstewart]

Seagate sells hard drives, and this gives them a cooler hard drive to sell. And there's enough intelligence in a typical disk controller to do most of the work here.

Intel sells chipsets, and their approach gives them a cooler motherboard chipset to sell, and hteir chipsets are flexible enough to add this kind of support.

Microsoft sells features that lock you in to Microsoft software, so if it's easy for them to support proprietary hard drives or motherboard chipsets or DRM chips that lock you in to buying and upgrading your Windows OS, that's what they'll do.

If you wanted to build a hybrid laptop for yourself, you could buy a $100 4GB USB flash memory stick, or a Compact-Flash-to-PCMCIA adapter (for mechanical safety) and some flash, install the operating system on the flash instead of the rotating disk, and use the rotating disk for backups and bulk storage. It'd be pretty easy with Linux (people fit mini-distros onto 128MB sticks, so why not on a bigger flash!), and you could probably get Windows to work instead. Seagate or Intel can do a more sophisticated job by working closely with MS to decide what types of data need to go in which caches - most Operating Systems today really only support two speeds of storage (RAM and disk-like) rather than knowing about three (RAM, Flash, Disk), (plus a few subtleties like swap space and networked filesystems), and really dealing with flash as a speed intermediate between RAM and disk will take a few years of research and some major OS reqork before it's done really really well, but even a naive version should be a big win.

For $20, you could do that again :-)

[billstewart]

256MB USB memory sticks are really cheap, and most modern PCs seem happy to boot from them, though of course you might want to spend $50 for 2 GB or $100-150 for 4GB. A Linux distro hefty enough to play Nethack should use up maybe 50MB :-)

The screen will still be burning battery, of course, but turning off the disks should help, and your data will load a lot faster because you won't have the latency of waiting for the disk to rotate or the head to seek (even if your memory stick has a lower transfer rate than the disk drives, you still should win.)

Definitely not cheaper

[billstewart]

1GB of RAM will cost you about $100 US. For that price you can get about 4GB of flash, either as USB2 or as CF.

The RAM is obviously faster, though it probably burns more power and generates more heat. On the other hand, you can install a lot of your operating system in a flash that big and avoid rotational and seek latency, which can make your machine a lot faster. And if you can get filesystem journalling to use it, that's a big big speed win.

If you've got a laptop, it's nice to use it to save to when you suspend/hibernate/etc.

The Real Advantage

[Siker]

The real advantage of this technology isn't necessarily the faster random access but the fact that you can keep the harddrive from spinning up. In every day normal operation there is always some little program that decides that now would be a great time to write a little bit of data to the harddrive. Maybe just a K or so for fun.

So even in a mostly idle system, there will be a bunch of little writes every so often. And since there are a couple of writes every minute there is no point in letting the harddrive spin down, even that even 100 writes only amounts to 100KB. You can't write-cache the data either because in order to give the harddrive time to spin down and really start saving power you'd need to cache writes for a very long time. That'd risk the integrity of the data since a sudden power loss would mean that data from a long period of time would be gone.

Enter flash on harddrive. Those 100KB won't make the 256MB flash even flinch. It'll take it. The power goes out ten minutes after the harddrive spun down? No problem. Flash mem's got your data covered.

In the end the harddrive might not need to spin up for hours if all you're doing is reading a very long E-book. Your room will be more quiet. Your battery will last longer. It'll be cool.