Samsung HDD Merges Flash, Conventional Storage

geekboxjockey points "This is a link to a story about a hybrid hard-drive technology from Samsung that involves the use of flash memory and conventional storage. A very interesting idea that could provide noticeable energy useage/speed improvements for HDD-based portable devices."

Comments

Suppose you had 10GB of primary memory...

[tomstdenis]

Yeah ... to be used for something called a ***CACHE***.

If you had 10GB of memory in a Linux/BSD box you'd get this "boost" too ...

As for saving costs by lowering failure ... let's hope they don't use cheap flash controllers. Of all the flash based mp3 players I've had [usually got for free with a purchase] most of them fail on a lengthy write or two...

More so let's hope we can still replace this hd+flash combo with a coventional HD.

I know for my Presario laptop [Compaq 2100CA] the replacement HD [Hitachi 60GB] is ==>$710 CAD== while a faster Samsung 40GB is $90 ...

So what i'm trying to say is ... less ass rapage please.

Tom

Re:Suppose you had 10GB of primary memory...

[Anonymous Coward]

I know for my Presario laptop [Compaq 2100CA] the replacement HD [Hitachi 60GB] is ==>$710 CAD== ... ...no, so what you're saying is that it's worth about $50 USD. Neat! I'll buy that.

Re:Suppose you had 10GB of primary memory...

[tomstdenis]

if by 50 you mean $564.97 you stupid inbred sister fucking wife beating jackass ... then yes, $50 USD.

duh hahaha you funny you make tired antiquainted joke.

Keep in mind there was a time our dollar was worth more. And also keep in mind the Canadian dollar doesn't go up. The USD goes down!

Tom

Re:Suppose you had 10GB of primary memory...

[TERdON]

The thing is, you wouldn't want to use volatile memory as write cache (even if you run Linux, some might cut the electricity), even if it works nice for reading. Having a flash memory integrated with the hard drive gives you the possibility to have all your data safely stored, and not staying in cache for 10+ minutes.

Re:Suppose you had 10GB of primary memory...

[MSZ]

Right, that thing looks like cache. I wonder how it will be operated on the OS level. Will it have internal processor to manage the cache and disk or will it be all done in the driver?

the replacement HD [Hitachi 60GB] is ==>$710 CAD==

You know, these HDDs tend to be standardized. Why not buy one and install yourself? While they tend to be twice as expensive as the normal desktop units, the kind of price you present is SICK.

What I did was simple: I keep the original HDD and put in decently priced (and fast (for a laptop)) Fujitsu drive. 60GB cost me about $200, interestingly, 40GB was only some $20 less. If I ever need to present the laptop for warranty repair, I'll put in the original drive. Just have to be careful not to scratch the screws too much.

Add $30 for an external USB box for the old drive, I now have also nice portable storage.

Re:Suppose you had 10GB of primary memory...

[tomstdenis]

I did get a replacement on my own. Hence the $90 for the 40GB drive...

The history went

1. Drive makes clicking noises
2. Tom sends it to Futureshop to get serviced
3. 56 days later I get laptop back
4. Laptop was not fixed, still clicks
5. Several months later it totally dies
6. Tom goes to local shop and buys replacement drive for $90.

Essentially the stores/manufacturers ALREADY rape us seven ways from sunday. This combo drive is just another way to potentially lock people down.

Tom

Re:Suppose you had 10GB of primary memory...

[Gentlewhisper]

The thing is, you wouldn't want to use volatile memory as write cache (even if you run Linux, some might cut the electricity), even if it works nice for reading. Having a flash memory integrated with the hard drive gives you the possibility to have all your data safely stored, and not staying in cache for 10+ minutes.

But the thing about flash memory is that they wear out after a certain number of write cycles. You wouldn't want to use that for your cache too!

Re:Suppose you had 10GB of primary memory...

[Anonymous Coward]

RTFA. It's a write cache. ... which provides capabilties an OS controlled volatile RAM cache cannot.

Re:Suppose you had 10GB of primary memory...

[TERdON]

But the thing about flash memory is that they wear out after a certain number of write cycles. You wouldn't want to use that for your cache too!

I do understand they aren't perfect. I also know there are ways to work around that, like checking the integrity of the flash (basically the same technique already used for disk drives, you could also use ECC technologies for example).

I however also know that as far as my hard drives go, the very same problem applies. They also have a limited lifetime (both disks of my athlon have died on me)... But you sure don't have a problem with storing things on disk, do you?

Re:Suppose you had 10GB of primary memory...

[MSZ]

Their reasoning seems to be:
has laptop = is rich = can pay triple = let's charge some extra

Yeah, I see this every time I go shopping for computer stuff.

Re:Suppose you had 10GB of primary memory...

[jdreyer]

This is basically Legato Prestoserve (introduced around 1989) packaged with a disk drive. Prestoserve used battery-backed NVRAM instead of flash. It was typically installed on servers, and its main purpose was not power conservation but rather accelerating an NFS server by committing writes to stable storage without waiting for the hard drive. But the idea is similar: a persistent disk cache.

Re:Suppose you had 10GB of primary memory...

[Anonymous Coward]

Awww....poor Canadians and their vicious inferiority complex. It's not our fault the NHL isn't having a season this year, ya Canuck fuck. And an American team would just win the Stanley Cup anyway, just like every other year. Go blow a moose, you worthless son of a whore.

Re:Suppose you had 10GB of primary memory...

[tomstdenis]

Inferiority complex? If anything what canadians have that sucks is apathy. We bitch at you americans but then share the same faults and don't fix it...

As for NHL, I personally hate pro hockey [and most sports that don't involve crashes or fire]. It's a scam and not worth the time and money.

Tom

Re:Suppose you had 10GB of primary memory...

[tomstdenis]

Hard disk write cycles and flash cycles are vastly different.

iirc. the average disk is rated for 10^10 - 10^11 [taken from Seagates average one bit error per 10^14 bits read/written] re-writes per sector while the best flash is around 10^5 - 10^6 [Intel strataflash].

Usually in a modern hard drive the motor will die before the platter becomes unusable.

Tom

Re:Suppose you had 10GB of primary memory...

[TERdON]

Yep, but that are that usual saying - there's lies, damned lies, and there's statistics. The fact is, most disks fail in a couple of year (or get so old you don't want to use them anymore, because they're too small/slow). They would actually be doing this, even if you just wrote the info to them once, and then let them run for a couple of years...

And - 10^5 rewrites should be enough for the envisioned usage, actually. As has been explained in other answers in this part of the thread, the flash would only be used when the hard drive has spinned down, and only for non-volatily storing write requests. A rough calculation shows, 10^5 rewrites would be enough, assuming one rewrite every quarterhour, to last for 2.85 years of use (it's not realistic they're continuous - at least in non-slashdot user scenarios). Higher disk activity than that would probably be a good motivation for spinning the disk up (and thereby not using the flash at all), lower would reduce the disk access frequency.

Also, the flash failing wouldn't really have to be a problem actually. Use some error-checking procedure (like is already being used on the disks themselves), mark bad blocks. In the end, ok, we won't used the fried flash, but we still have a "usual" disk to use.

Swap File

[eXzite]

Will the OS have a way of selectively writing to the flash? What about swap files, etc, which will change all the time?

This is great stuff though, swap files aside, most people could probably do everything they ever need from their laptop within 1Gig of flash, during a single work session.

Re:Swap File

[Eric604]

That doesn't really matter. It's about preventing the hdd to spin up. If that means storing a swapfile page on flash, then it has to be so.

Re:Swap File

[Anonymous Coward]

But that's undesirable insofar as it induces early hardware failure due to exceeding the read/write cycles of the Flash.

Re:Swap File

[Janitor]

Thats where the 10GB of primary memory comes it, so you won't need to swap. They thought of everything.

Re:Swap File

[Eric604]

The OS should minimize swapping. If you want to prevent the hdd from spinning up you HAVE to use the flash memory for the swapfile, otherwise.... well, the hdd spins up, it's not that difficult.

Re:Swap File

[MSZ]

For Linux, sure.
For Windows, not unless they replace the memory management. It normally uses swap even when there's plenty of free memory and this is supposed to be a feature, not a design bug...

A gimmick

[Saven Marek]

flash memory like this in a hard drive is a gimmick, I think samsung are hoping nobody realises how few the write times are on flash memory, so you'll need to regularly replace the limited write time flash memory when it's worn out fairly often

So what happens when trying to detect when the flash memory has been written to too many times? afaik this isn't easily done, so you end up dumping broken data to the disk until you notice "whooops my spreadsheet suddenly doesn't work that I need in 30 minutes or the boss will have my ass".

Re:A gimmick

[drinkypoo]

So what happens when trying to detect when the flash memory has been written to too many times? afaik this isn't easily done, so you end up dumping broken data to the disk until you notice "whooops my spreadsheet suddenly doesn't work that I need in 30 minutes or the boss will have my ass".

you could be right, but I doubt it. What could be simpler than verifying all writes? of course, it's slow, so you might want to do it in a delayed fashion; eg have some memory in there, write to ram and to flash, do a comparison...

Re:A gimmick

[Cruithne]

So what happens when trying to detect when the flash memory has been written to too many times?

Actually, in most newer flash-based storage devices, this is already accounted for. Basically, the data is verified by attempting to read back the data - if it reads, you know you're fine, as the "space" in memory will only become "worn out" on a write.

If a "space" fails verification, it is added to a list of known bad sectors - exactly how IDE drives have functioned for the past decade or so.

Re:A gimmick

[Anonymous Coward]

The write errors come up slowly so you use an ECC correction algorithm to correct errors and when it gets higher than 2 or 3 bits you no longer use that flash memory page and move the data somewhere else in flash.

Re:A gimmick

[rsmith-mac]

This is more of a "may be a problem" than a "will be a problem" kind of event. It's all dependent on how much data is actually moved, and how the Flash memory is configured. For example, if we have a 1GB buffer, then at a 100K-write life, we're talking about 100TB of data that can be transfered before the Flash memory wears out. The question is how does that 100TB number compare to the amount of data actually moved in a normal drive's lifetime?

Re:A gimmick

[St. Arbirix]

They talk in the article about Longhorn using 64-bit memory addressing so you'll be getting gobs of main memory. When your page table is 10GB like they mention in the article then there would be little to no reason for another paging system inside the harddrive. As much as I like modularity this is just silly.

It's definitely a gimmick, and a nearly useless one when everyone's upgraded to their 64-bit systems. Do you want your memory cached on something connected to the FSB or do you want it done out on a device? I don't even understand why they're trying to get into this game.

It's like making a pencil that never wears down but putting the same old erasers on it.

Re:A gimmick

[GigsVT]

I guess it depends on what normal is. It takes a few days to move 2TB of data across a network usually. If you were thrashing swap constantly, 100TB could be a month. For desktop use, I'd say it would be at least a few years.

Re:A gimmick

[oc255]

Gimmick? Hmm...

This is just electronic writes. Those who have worked with a high-performance SAN like the Hitachi 9900, Sun 6920, know that electronic writes is where all the performance comes from. When our SAN's 4gig of cache goes offline, my DBAs come running and everyone complains about terrible write/read speeds.

Electronic writes (in a good amount) means that the data flies into memory and later on the disk system pumps the data out to the disk platters. Netapp is really great at doing this kind of operation. Talk to someone with a Netapp and they'll tell you that you can watch all the activity lights on an array light up in a fury of activity with the Netapp head pushes from cache->disk. It's really fantastic. Of course, it's really expensive too.

I'm not saying that Samsung's disks are going to be enterprise-class. But I'm hoping that this cache thing catches on to make SATA2 (or whatever) based arrays feasible for desktops. Imagine a chain of large-cache drives on desktops, large-cache single drives on laptops ... For example, if you had 128MB of cache per drive, then an array of 8 disks would have 1gig of cache. That would be so great for the power-user. The Sun T4 bricks cost $50k and they have 2gb of cache per controller. 2gb of electronic writes/read = amazing performance.

I can't wait to learn more about the quality and speed of the disk cache. It's what I've always wanted to see in consumer grade disks.

Re:A gimmick

[zyridium]

There is still a good reason: It is non-volatile.

Re:A gimmick

[mp3phish]

YOu don't understand the fundamental difference between this and a "cache"

This is NOT a cache. It is a permanent storage area on the hard drive which does not require the hard drive to spin up. I know one thing, in windows XP, my laptop hard drive spins up every 10 minutes because XP likes to do tons of shit even when i'm not using it. All it does is write 1 or 2K onto the disk, and for that it spins the damn drive up... every time. Witht his embedded flash memory it can write to it, and only after a long period of time, when those 1K writes add up to 128MB.. then the drive will spin up and copy all that flash ram into the hard drive... and the cycle starts over..

You see, most people who are using their drive to browse the web, work on a word file, IM chat, webcam chat, do email, etc etc... they aren't using their hard drive more than 128MB.. probably at most they are reading/writing to and from the drive in less than 64mb in a few hours time (provided you already have the applications loaded and you don't waste disk space on web cache) You save yoru word file every 5 minutes, but why spin up the disk and keep it running that whole time when all you are doing is writing out a few K every once in a while? You send and recieve a few emails, but why spin up your disk just to use it to send an email?

The point here is that battery life will significantly improve (probably by about double), and hard drive MTBF will significantly improve (probably more than double) in a laptop if you use this small flash PERMANENT (not cache) storage while your disk is spun down...

(notice that the flash memory is never used while the disk is spun up.. it is only used while the disk is spun down and it wants to try not to wake it up).

Re:A gimmick

[mp3phish]

While your dreaming sounds wonderful, and I wish it were so, this particular circumstance isn't a cache. It is permanent storage built for one thing and one thing only: to prevent the drive from spinning up and still store data permanently (up to a certain size)

This flash memory will be slower than the hard drive's native writing. It will definately not work like a cache. It would destroy the flash memory in a matter of months, and would be much slower, if you used it like that.

The purpose of this design is so when the drive goes into sleep mode due to inactivity, the OS can still permanently write a file to the disk without waking it up. If you save your word file which is 2kilobytes, you send it to the disk, and the disk will spin up on a conventional system. On this new system it will write it to the flash storage.

Once you write 128MB of data to the disk, the disk must wake up and copy it over to the platters.. once the period of inactivity happens again(1 minute to several hours depending on your power management settings), the disk will spin down and the cycle will repeat.

But please don't confuse this with a cache. It doesn't work like a cache. It doesn't perform like a cache. And if you used it like a cache you would have a brick in as little as a month.

Re:A gimmick

[St. Arbirix]

I know one thing, in windows XP, my laptop hard drive spins up every 10 minutes because XP likes to do tons of shit even when i'm not using it. All it does is write 1 or 2K onto the disk, and for that it spins the damn drive up... every time.

What part of 64-bit addressing did you not understand? Or was it the virtual memory part, cause I know that sometimes gets people?

Of course it's only writing a few kB to disk, that's the page size that your virtual memory manager uses. What it means is that your real memory is full and something had to be written out to virtual memory on the harddrive before it was overwritten in real memory. This sounds like you'd be storing your virtual memory in flash instead of on a plate, which is pretty useful, but that use is severely diminished when you no longer have limits on how much RAM your computer can handle which will effectively remove the nead for virtual memory on disk that's getting constantly written to.

The silly thing about the article is that they're touting this new Samsung drive while at the same time touting Longhorn's capability (as a 64-bit OS) to make the Samsung feature extraneous.

Re:A gimmick

[mp3phish]

The silly thing about the article is that they're touting this new Samsung drive while at the same time touting Longhorn's capability

They are touting longhorn's ability because they are hinting that they will be decreasing the swapyness of the system when you use lots of RAM. They will also try to decrease having to load and unload libraries n stuff you need the most... They are basically saying, that with this new technology on the hard drive, along with tweaks from the OS, that you can pretty much live with just 128MB of permanent storage MOST of the timem.. and the times you need more the disk will spin up and give you access to it.

You are making it sound like 64Bit processors will not add more RAM to your system. You are making it sound like it is a useless improvement. You are making it sound like no matter WHAT, swap will destroy the benefits of this technology. You know what? You are wrong. It is a matter of tweaking the VM to help prevent the system from swapping in and out excessively while in a lower power state. It is just a matter of time before this is implemented on the OS level (in linux, macOS, and longhorn)

Re:A gimmick

[ModMeFlamebait]

It is just a matter of time before this is implemented on the OS level (in linux, macOS, and longhorn)

yeah, /proc/sys/vm/laptop_mode and associated acpi scripts are sure a long way into the future. not. it has been present in 2.4 linux kernels for quite a time. when longhorn comes out, it'll be *years* behind schedule

Re:A gimmick

[gonzo_bozo]

Solid state storage will give a second life to lots of computer that would be discarded otherwise. The one thing that slows down my computer at the moment is my hard drive reading files.

Re:A gimmick

[gonzo_bozo]

From wikipedia Flash_memory : "Its endurance is 10,000 to 1,000,000 erase cycles".

This is much more than I need for an overwhelming majority of files written on my disk.

Sectors that are accessed regularly and written seldomly get eventually written to flash. If this ever changes with time, these sectors are moved back to the magnetic medium. MTBF will most likely be due to the mechanical part of the drive.

Running hdparm -tT on my system:

Timing cached reads: 1156 MB in 2.00 seconds = 578.09 MB/sec

Timing buffered disk reads: 60 MB in 3.17 seconds = 18.95 MB/sec

That's 30.5 times faster.

I want my hybrid disk. Do you?

Re:A gimmick

[Anonymous Coward]

"From wikipedia Flash_memory [wikipedia.org] : "Its endurance is 10,000 to 1,000,000 erase cycles". "

more than enough for me.

as long as i can save naughty-allie once to the hd i never need to change this data or erase it.

as long as i can access/read it a couple of times a day im satisfied.

Re:A gimmick

[Barlo_Mung_42]

When they demoed this at WinHEC last week I asked about this. They said it would take about 40 years of typical use to degrade completely. Even then it degrades gracefully so the cache just shrinks until you end up with a normal hard drive (assuming it lasts that long).

Re:A gimmick

[mp3phish]

it has been present in 2.4 linux kernels for quite a time. when longhorn comes out, it'll be *years* behind schedule

Yea.. but you have to admit that the linux kernel has a long way to go for quality acpi support.

You might say this feature is in linux, and has been for a while. That doesn't change anything I stated in the above post. It only says that Linux has had this feature. I never said it didn't. And my statements aren't based on Linux not having it. In fact, I don't care for windows. It just so happens that someone posted misinformation to slashdot about longhorn and I corrected it. It doesn't make me anti-Linux or pro-Windows.

software implementation...

[Nova1313]

Isn't this already implemented in software in some form. As memory cache of sorts. I understand that with memory if you loose power you loose data so this just seems to act as a bridge. What happens if the power goes out and data has yet to be written but is in flash. Is what is in there automatically committed to the disk on the next power up? The article doesn't go into much detail unfortunately. It seems like a good idea if implemented properly but for me everytime I launch a new copy of explorer on a windows box it has to spin up the disk. So unless you keep the most frequently used applications in memory I'm not sure this will cut down on writing except for with small data files being saved.

Re:software implementation...

[Anonymous Coward]

lose = to be deprived of something
loose = your mom

learn it, love it

Re:software implementation...

[hedora]

This is a really old idea. Initially, the idea was to run some system memory or drive cache off of a battery. Then, a portion of memory would remain persistant across reboots. The memory would act as the flash memory in the article does, except that it would be normal system memory, so it would be fast, but you would have to worry about battery failure. I'd like to see something like this in the linux kernel. You could imagine having it support spooling reads and writes from any type of block device to any other block device. Then, you could just plug flash into your system, or you could use battery backed up ram. If you want good redundancy, you could use a fast raid 1 setup and slowly spool data to raid 5. There are quite a few pre-existing products that do all of these things already, but they are generally expensive, or not widely available. I don't see why any of it couldn't be done in the operating system with standard components, though...

Re:software implementation...

[mp3phish]

No this is not implemented in software yet. This isn't a cache. It is a storage buffer for a hard drive. It is significantly slower than a memory cache. It is even slower than the drive platters themselves. It is simply slow flash memory designed for one purpose only. That is to have a place to write to while the disk is spun down.

Cache does not provide this function because cache is volatile and if you write to it, you will lose it unless it writes it to the platter. Whith this system, you can write to it, and it is not lost with a power failure.

If you were to use this flash memory like a cache your memory would fail in a matter of months because flash memory of this sort isn't designed for continuous reading/writing to on an every day basis.

Nah...

[ArbiterOne]

It won't really take off unless they release an amusing Flash video involving dancing flash memory-HDD pairs singing about the joys of dual storage.

Hardvaporware ?

[alexhs]

> The Hybrid Hard Drive, developed by Samsung and Microsoft, is meant for mobile PCs running Longhorn, the next version of the Windows operating system.

So the other hard drive manufacturers will have a loooong time to do the same...

Re:Hardvaporware ?

[Anonymous Coward]

It is jsut a facy thing only, read/write durability is less for flash drive compared to hard drive

Technology to reduce 10% of power consumption is not looking good or great to me when there are great impovements coming in battery technology.

Re:Hardvaporware ?

[senatorpjt]

Not to mention that by the time Longhorn comes out, I'll probably already have solid-state holographic storage on my Powerbook. :)

Re:Hardvaporware ?

[mp3phish]

You imply that a powerbook made by Apple Computer would somehow implement futuristic types of storage faster than other platforms. You might need a history lesson:

Apple was the last mainstream manufacturer in the industry to implement USB2.0 into their systems for use with Hi Speed USB flash memory devices.

Apple was the last mainstream manufacturer in the industry to install SATA drives into their systems.

Apple was the last mainstream manufacturer in the industry to install DVD burners into their systems.

Apple was the last mainstream manufacturer in the industry to move from SCSI to IDE for consumer systems.

I highly doubt Apple will be the first manufacturer to implement holographic storage into their laptops. As they are usually the last manufacturer to move to new types of storage technology in their products.

Re:Hardvaporware ?

[TylerL82]

...of course, you could also say...

Apple was the first mainstream manufacturer in the industry to implement USB into their systems (iMac, May/August 1998).

Apple was the first mainstream manufacturer in the industry to implement FireWire/IEEE1394 into their systems. (January 1999, Blue and White Power Mac G3)

Apple was the first mainstream manufacturer in the industry to implement DVD burners into their systems (the original SuperDrive, DVR-103, was introduced in January 2001 alongside the Pioneer DVR-A03. The drive was the first commercially-available burner for DVD-R General discs).

Re:Hardvaporware ?

[evilviper]

Apple was the first mainstream manufacturer in the industry to implement USB into their systems (iMac, May/August 1998).

No, you're COMPLETELY wrong there. USB had been installed in PCs for a LONG time before Apple's iMac. I've personally got a DEC PentiumPro system with USB ports, from 1995 IIRC. What Apple actually did, was popularize them, by forcing their customers to use only USB peripherals. I'm still using it as my firewall, loaded up with 192MB of old SIMMs I don't have any other use for...

The ironic thing about Apple pushing USB was, it undermined their attempts to popularize their own firewire standard. They pushed USB for slow devices, and firewire for fast devices, and instead, USB took over for almost all devices, fast and slow.

Re:Hardvaporware ?

[Anonymous Coward]

umm, you completely missed his point.

Apple was the first mainstream manufacturer in the industry to implement USB into their systems (iMac, May/August 1998)

Lets re-read that statement: Mainstream manufacturer. As in, the kind of company that was shipping boxes with USB ports & USB keyboards / mice.

Yes, Wintel PCs had USB on board. I had a 430HX motherboard with USB on board it, long before the iMac. Guess what? You couldn't use it without a special version of Windows 95 (which didn't even exist when I bought my motherboard..) The Win95C USB implementation was lacking in a few departments (hell, it took till Win98SE for USB Mass Storage to make it to the PC).

So, getting back on topic: Apple shipped systems with USB on board, an OS that could use USB and actual USB hardware. That's more than any PC maker that puts boxes on shelves could claim. Apple actually *used* USB, creating a market for USB periphs & showing the 'average' consumer what it was.

In all fairness, blame Microsoft for wasting time with USB. The hardware was on motherboards for years before MS got their act together.

And blame Intel for the Firewire thing. Apple plugged USB for the slow, Firewire for the fast. Intel decided they would *not* put 1394 on their motherboards (despite telling everyone they would). In fact, the last Intel-based motherboard I bought (2 yrs ago) still didn't have 1394 on board. Any Intel chipset based motherboards you find with 1394, you'll find it's not a part of the south bridge, it's on the PCI bus.

Anyway.. Just needed to set the record straight. I use Intel, AMD and Apple every day, each has their own use. Just don't short-change Apple's influence on the industry.

Re:Hardvaporware ?

[Anonymous Coward]

Apple was the last mainstream manufacturer in the industry to install DVD burners into their systems.

This is the biggest piece of tripe I've ever read. Apple was the first mainstream manufacturer to ship a system capable of turning DV into DVD off the shelf. They did it before HP, Compaq, etc.

Re:Hardvaporware ?

[Anonymous Coward]

In 1998, was Apple really all that much more mainstream than DEC?

Re:Hardvaporware ?

[Anonymous Coward]

More mainstream than DEC? Probably not. Apple has always had more exposure than market share (their market share remains low, but everyone knows who Apple is and what they're up to). That is to say, Apple products are on a lot of store shelves, the same can't be said for DEC (then or now). However, that doesn't make Apple more mainstream, just more exposed (as I see it).

However, it's a little misleading to suggest DEC beat Apple to USB by 3 years (1995, USB-enabled PPro. 1998, iMac) when Microsoft wasn't anywhere near providing a USB-enabled operating system in 1995.

Apple shipped a consumer system that was 100% USB in 1998. In my mind, that's a major accomplishment. There were no all-USB computers besides iMacs at that point.

Re:Hardvaporware ?

[evilviper]

Lets re-read that statement: Mainstream manufacturer. As in, the kind of company that was shipping boxes with USB ports & USB keyboards / mice.

That's absolutely ridiculous. You're just redefining terms to suit yourself. Mainstream manufacturer means a big, well-known company. It has nothing to do with how well supported anything is.

You couldn't use it without a special version of Windows 95

Well, that's not true, either. There were USB drivers available before Win95 OSR2. USB support in the OS just made USB drivers easier. Companies like IOMEGA just wrote their USB drivers from scratch before OSR2 came out. Hell, they still provide USB drivers for NT4, which never got USB support.

And despite your claims, USB support in 95 OSR2 was perfectly fine. Sure, UMASS support didn't come around for a while, but it was never necessary, just a bit more convenient. And you're wrong about the version, too. 98b didn't have UMASS, it was ME and 2000 that were the first to get UMASS.

Besides that, USB drivers aren't even necessary. USB keyboards/mice appear as PS/2 devices to the OS, so any OS could use USB keyboards/mice as soon as the hardware was available.

Apple shipped systems with USB on board, an OS that could use USB and actual USB hardware.

If you had bothered to read my entire post, rather than just the first sentence, you'd have seen that I explained in detail exactly what Apple did. They FORCED their customers to buy USB devices, giving them no other options, which greatly promoted USB devices, and undermined the firewire market.

Apple plugged USB for the slow, Firewire for the fast.

Yes they did, but they shouldn't have. Their support of USB for slow devices allowed it to get a strong hold on the market, and undermine firewire. If Apple had made firewire a bit better, and used it for both fast and slow devices, we'd all have firewire everything, and USB would have died off, as computer users had already spoken, and said they did not like USB.

FROM Here to Eternity

[Doc Ruby]

How about booting from the HD, then "caching" the computed startup image in the FROM? At shutdown, store kernel/OS variables in a table. At next boot, just suck in the image from FROM, and update runtime changes (clocks, counters, etc) from the table. Corrupt images get dropped by rebooting from HD when necessary. It's like notebook "hibernate", but stores the "clean" initial boot state instead of the (possibly corrupt) final OS state. Linux's initrd boot ramdisk phase offers a golden opportunity to just restart from the image cached in quick FROM. If Samsung patched the bootloader, it could sell a lot of these drives. I'd pay as much for a 100GB platter-only drive as I would for a 40GB boot hybrid drive.

Re:FROM Here to Eternity

[Anonymous Coward]

that makes no sense.

The variables that you are saving are just whatever they ended up at after initializing the devices.

You can't just skip initializing the devices...

or someone would have just saved the image and reloaded on boot years ago.

Re:FROM Here to Eternity

[Doc Ruby]

It makes perfect sense, if not to you. Most of the image is the loaded code, selected by boot-time options. Some of it is the state of the variables after init, including devices, which is the same after every reboot. And some is the state of the variables, including devices, which varies by external factors (network, sensors) and by the last state before shutdown. That last recomputation is a tiny fraction of current boot time, which is largely the disk IO time as tiny bits of the OS and configs are read depending on boot calculations. In your sense, the v2.6.12 version of the kernel is impossible, otherwise someone would have just coded it years ago.

Re:FROM Here to Eternity

[Anonymous Coward]

Three words: peripheral hardware initialization. Tends to be more complicated for the general case of PC's than the manufacturer specified laptop.

Re:FROM Here to Eternity

[Doc Ruby]

Yes, but that init is just the majority of the state needed to be recreated at boot, which is a minority of the boot time. Still lots of gains from caching the rest, rather than recomputing it every time.

Read/Write times

[blueadept1]

Correct me if I'm wrong, but wouldn't a flash-based hard drive have quite a bit (~2x) slower write and read times? Perhaps if it was built into the architecture to be 'closer' to the processer such as RAM is.

This seems like they are just trying to turn RAM into a permanent storage device.

I really don't see any advantages to this as it is. Even the prices would be higher.

well, then, not new...

[fraxinus-tree]

AFAIK, modern filesystems can store their journal on a different device. You just need to change journal flushing logic.

Déjà entendu ?

[alexhs]

"Let's suppose you had 10GB of primary memory--probably everything that you do could fit in memory," Allchin (Microsoft Windows chief) said.

10 GiB ought to be enough for everyone...

Re:Déjà entendu ?

[Anonymous Coward]

Well that's all well and good. But should I use ketchup on a sausage sandwich?

Re:Déjà entendu ?

[ScrewMaster]

No ... mayonnaise, and only Hellman's Real Mayonnaise or you're just wasting your time.

Actually this story isn't entirely accurate

[gru3hunt3r]

There have been several vendors of Flash Based hard disks for a while. This is the first hybrid flash+magnetic drive -- and even this isn't all that different of an idea than say a Compaq smart array controller with battery backed write cache which used NVRAM to store data. It's innovative and i'd definitely buy a laptop that had it.

I think many slashdotters will miss the big picture. This is mostly a power saving utility -- and it could offer performance gains assuming the files you use are available on the flash and the drive doesn't need to be spun up. (Of course when the drive DOES need to get spun up, plan on having a *really* long access time so I think this will be negligble). Buy basically it means you can leave auto-save on Microsoft Word enabled and not drain your battery.

BUT since we're on the subject i'm a huge fan of flash only drives, they have several special applications because of their access times (in nanoseconds instead of milliseconds), extremely reliable (no moving parts, read/write cycles in the billions + ECC checking) and high bandwith they are NOT ideal for situations such as swap (JUST BUY MORE RAM IT'S CHEAPER AND FASTER!!) but instead they are perfect for situations were you need persistent storage of highly accessible files e.g. binlogs on a database.

You can easily bump up the performance of MySQL or Oracle using one of these drives for A LOT less

There is a company called BitMicro http://www.bitmicro.com/ which produces ATA and SCSI, and Fibre Channel flash only hard disks.
Using a flash only drive you will get a dramatic performance bump in any transaction database by storing the transaction files on the database.

Re:Actually this story isn't entirely accurate

[Jeff DeMaagd]

Flash memory has have a limited number of writes, hard drives have a lot more writes by orders of magnitudes. Flash memory writing can also be very slow.

I think what you mean is battery backed up memory, which I think is ordinary DRAM chips with battery backed memory controller to keep the memory refreshed.

Re:Actually this story isn't entirely accurate

[DigiShaman]

Also check out http://www.m-sys.com/

They do offer solid state drives in ATA and SCSI format too among other such products.

Re:Actually this story isn't entirely accurate

[the frizz]

gru3hunt3r's says: Compaq smart array controller with battery backed write cache which used NVRAM to store data.

I can certainly back him up on that. We have them at work. Last week the disk performance went down a third on one disk array. On the BIOS bootup screen did we saw an error message that effectively said the battery had failed on the disk controller so it had to wait for each write to go to disk rather just return ok when it was written to cache. Things recovered when the battery was replaced.

And given current USB stick (which use flash) prices of $80 per GB, using flash RAM as a cache for disk seems like a good idea compared to building the whole disk out of flash.

Re:Actually this story isn't entirely accurate

[ScrewMaster]

Yes, I've burned up several flash sticks because I write to them constantly at work (usually 60-100 Mb application setup sets) so I can attest to the limited number of write cycles. Ordinary DRAM parts aren't suitable for battery backup unless you include refresh logic and in any event they always draw substantial amounts of power. The solid-state drives I've used have CMOS static RAM on them. CMOS logic gates don't draw much current unless they're in transition, which makes them ideal candidates for long-term battery standby. Being true RAM (instead of flash, which is basically just a modified EEPROM anyway) there's no particular limit on the number of reads or writes. On the other hand, they don't have the bit density of dynamic RAM, and the cost per bit is much higher. But they still are excellent for industrial applications where you want reliability (harsh environments of one kind or another) or where a system has to boot very rapidly.

Re:Actually this story isn't entirely accurate

[div_2n]

You can easily bump up the performance of MySQL or Oracle using one of these drives for A LOT less

From TFA, "The Hybrid Hard Drive, developed by Samsung and Microsoft, is meant for mobile PCs"

If you are doing anything on a laptop with a database that might require a "performance bump" as you put it, then I would suggest getting at a bare minimum a hefty workstation.

I don't think too many companies would run a production database on a laptop.

Re:Actually this story isn't entirely accurate

[gru3hunt3r]

div_2n? Did you ride the short bus to school or what?

I said:
I'd definitely put one of these in my laptop since it would save on power.

I didn't say use one of these drives for a server -- I said use a PURE flash drive to store binglogs -- but the article touts the drives as some sort of new invention and I was merely pointing out that using flash hard disks isn't new.

But since we're on the subject -- why not run one of these drives in a server? especially any server which does frequent writes and then reads of data --- seems like it'd give you a performance bump there as well (think temp files).

Re:Actually this story isn't entirely accurate

[SuperQ]

flash drives are great.. what does bitmicro have for price/gb?

Re:Actually this story isn't entirely accurate

[bergeron76]

You can easily bump up the performance of MySQL or Oracle using one of these drives for A LOT less

I too am very concerned about getting extended battery life while running Oracle 8i Server on my laptop (with mysql).

As for BitMicro, I received a quote from them for a 1GB flash HD, and it was around $990 USD.

No thanks. I'd rather use a CFcard with an IDE adapter. Granted, it's not quite the same; but it's certainly a more reasonable value.

Re:Actually this story isn't entirely accurate

[gru3hunt3r]

No no no..
A CF-Card is *WAY* too slow. Plus residential memory sticks aren't designed to be durable enough to withstand the normal read/write operations required by a database, nor do they typically have the bandwidth (so even though the seek time is very fast, the bandwidth on most types of residential memory cards is pathetic). DO NOT USE CF-CARDS IN YOUR SERVERS THEY WILL DIE AND YOU WILL BE SCREWED.

Yes -- I agree. BitMicro stuff is expensive, though if you are doing a quantity purchase you can get a better price. One of their primary customers is the military (they use 'em in Missles) so I think they are accustomed to charging too much. :-)

If you can't do a quantity purchase then talk to some other friends who are admins and you all go in together and they'll cut you a deal.

But even at at retail a 1gb drive (which is WAY MORE THAN MOST PEOPLE NEED FOR THEIR BINLOGS) should give you a 40-50% performance boost on updates/inserts/deletes for MySQL -- even more if you use replication. You tell me where you're gonna put a $1000 and get those numbers.

They also claim to have a special patented method for how they write to the drive to spread the writes around so that the drive doesn't burn up and so they claim it has an equivalent MTBF than a hard disk (I think most drives around 30 years these days, and they claim 25 years on their stuff). ..

Anyway.. bottom line: don't use a CF-Card, I bet most ATA-2 IDE drives could beat the bandwidth on one of those.

History repeats itself...

[banuk]

Is is just me or does: "Let's suppose you had 10GB of primary memory--probably everything that you do could fit in memory," - Allchin (FTA).

sound similar to: No one will need more than 640 kb of memory for a personal computer. - BIll Gates (IIRC)

Re:History repeats itself...

[alexhs]

> Yeah, and you are repeating a poster from 5 minutes ago.

Haaa, but... there is a difference !
He is getting an insightful mod while I'm getting a funny !

Flash and harddrives

[karvind]

CeBIT 2005 had demonstration of flash only hard drives. Since flash memory is considerably more expensive than magnetic mass storage - a hybrid approach is a better compromise.

Also from WinHEC, samsung is not the only player. The disk will be manufactured initially by Samsung, Hitachi and Seagate, and other manufacturers will be announced later.

More details on Samsung's OneNAND hybrid technology:

OneNAND Flash memory has been incorporated into the design of Microsoft Corp.'s prototype Hybrid Hard Drive (HHD), the first fully functional disk drive to combine NAND-based Flash with rotating storage media.

The hybrid hard drive prototype uses 1 Gigabit OneNAND(TM) Flash as both the write buffer and boot buffer. In the hybrid write mode, the mechanical drive is spun down for the majority of the time, while data is written to the Flash write buffer. When the write buffer is filled, the rotating drive spins and the data from the write buffer is written to the hard drive.

The hybrid drive saves power by keeping the spindle motor in idle mode almost all the time, while the operating system writes to the OneNAND write buffer. Moreover, by using OneNAND Flash with hard disk drive technology, disk drive performance is not compromised relative to conventional disk drives. This is due, in large part, to OneNAND's ultra-fast read speeds, which can be fully leveraged during the flushing of the contents of OneNAND's write buffer to the rotating drive. In addition, since the Samsung hybrid disk drive operates at a lower temperature than traditional rotating media, it greatly reduces the possibility of shock and impact damage, improving the overall reliability of the disk subsystem.

While the cost of hybrid disk drives may slightly increase with the addition of OneNAND, any increase will be mitigated by several factors, including lower maintenance costs, 95 percent power savings when the disk is not spinning, faster boot time and substantially increased reliability. All of these changes are crucial to the ever increasing needs of today's mobile customer, making it likely that hybrid hard drive technology will enjoy rapid market adoption.

What's so innovative about it?

[DrLex]

So if I get it right, this is just a drive with a very large, albeit slow, memory cache. They expect the users to address their data in separate chunks of 128MB. As soon as you go outside of this chunk you'll have to spin up the drive to read the file -- which will, of course, reduce the responsiveness of the system. Moreover, drives don't only wear out due to the disk spinning. Every spin-up and spin-down cycle causes additional wear, so I doubt this idea will reduce the failure rate for laptop disks. Actually this article looks more like publicity for Longhorn...

Re:What's so innovative about it?

[uberdave]

Keeping a disk spinning is a major power consumer. This type of system could significantly extend battery life.

Re:What's so innovative about it?

[Anonymous Coward]

it's because the harddrive cache is flash based hence NON VOLATILE and can be optimised to cache files before the OS has even booted thats whats so innovative.

So writes get cached to the flash, if a BSOD happens then cache contents aren't lossed.

Power savings?

[PolyDwarf]

The article says that they're thinking the higher cost of the drives will be offset by maintenance/power savings.

Me, as a laptop buyer, doesn't give a rip about either of those. Power? So what, I fill up on power at the coffee shop, at my office, etc, if I'm concerned about paying for it. It's virtually impossible to do maintenance on a laptop, other than wrap it in a box and send it in, in which case if it's a personal machine, I just use my desktop, and if it's a work machine, I still get my salary.

The only people it seems this would help are self-employed people, who have no other computer, who live in the boonies.. And how any of those are there?

Re:Power savings?

[Anonymous Coward]

a laptop buyer who doesnt care about power ?
LMFAO

Re:Power savings?

[Anonymous Coward]

I don't suppose you've ever tried using your laptop on an international flight. Or heaven forbid you actually using your portable computer when you're not right next to a wall outlet.

Maintenance

[XanC]

As you point out, you're not doing the maintenance yourself. But you're sending it in for it to be done.

If the manufacturer can budget less money for maintenance, he can budget more for features or lower the price of the laptop.

Re:Maintenance

[mp3phish]

Exactly.. Likewise your employer can have less downtime therefore getting more work for the salary they are paying you...

Not to mention the most important issue: the power savings you get while on a battery..

It sounds like the parent poster is just trying to be grumpy and think that because he never uses a battery, and he doesn't know how to replace a hard drive in a laptop (actually, most laptop mfg's consider HDD's as end user servicable nowadays) and that if it goes down on him while he is at work he will still get paid... that none of this matters to him...

Hate to break it to him but the world doesn't revolve around one person.

Faster session recovery times

[alexhs]

All Laptop Data in Main Memory?

So when you're going out of battery, it's like your session(*) never was...

(*) and the work you did, too

useful optimization

[ArbitraryConstant]

I believe this won't be directly used as cache (as the limited write cycles of flash memory would make this impossible), but it will provide an area where relatively static information (like the kernel, libraries, etc) can be stored and accessed without spinning the drive up. Obviously the OS needs to get involved because only it is in a position to know what files should be placed in this cache.

If MRAM ever becomes economical, it might be useful as a non-volatile general-purpose cache. That would be handy because it would reduce the risk of corruption on power loss.

Re:useful optimization

[Anonymous Coward]

Yeah, great idea. NOT

Like the world really needs a memory that is in cost/performance in between ram and magnetic storage?

That nieche product exists. It is called bubble memory.

Very short lifetime....

[yani]

It seems that most people are forgetting a very important fact - flash memory has a limited number of writes. For normal usage (e.g. sd card/usb memory key) you will never encounter these, but as soon as you do something like this...

This just doesn't make sense to me, instead of caching in system memory (how the power savings are done in linux laptop-mode) they are caching to flash which is slower, the only advantage is the data won't be lost in flash.

However what about swap? Does the flash cache distinguish between swapfile writes/normal data? If not you just bought a hard drive with a very short lifetime. Even if it only uses flash for normal data writes, this drive will have a considerably shorter lifetime.

Re:Very short lifetime....

[Anonymous Coward]

Samsung is the largest manufacturer of memory chips in the world. I wonder if they might have an employee or two who has thought about this? :)

Re:Very short lifetime....

[ajlitt]

yes and no. Flash has a limited number of writes per cell. However, like a hard disk, most flash has backup blocks that can be enabled and disabled as needed when the good blocks go bad. Also, flash controllers (i.e. the chip that attaches a NAND chip to the USB bus in your flash keychain) implement something called wear leveling, whereby the flash writes occur evenly throughout the memory device even though the host may be overwriting one block more frequently. Think of it as intentional fragmentation.

Also, flash devices don't write the contents presented to the data pins directly to the flash cells. Data to be flashed is copied into an on-chip SRAM buffer (4-128k is common), and a command to write the block is given. The flash chip handles the block write on its own, and signals the host when it's done. Before the buffer is overwritten, a verify command can optionally be given, causing the chip to go off again on its own and compare the data. The compare/verify is not only autonomous, but much faster than the erase/write operation, so verifying that your data got written properly is not a big performance hit. If the verify failed, well, you can mark that block bad and swap in a spare.

All this means that, in the worst case, your flash buffer for your hard disk will slowly shrink over time, but your data will still be safe because it will eventually get to the disk anyway the next time it spins up.

The advantage of this over using a large RAM disk cache (which works well under Linux for me as well) is that when the disk DOES need to be written to by the OS, the flash sees the write instead of having to spin up the disk to do the write, which is a very power-costly operation. Flash can go into a write state and back to sleep almost instantaneously, with no significant ramp-up power requirements.

BTW, in this day and age, RAM is so cheap and plentiful that in a well designed laptop system you should have enough RAM that swapping to disk is unnecessary.

Re:Very short lifetime....

[arodland]

On that last topic: the "laptop mode" feature in the linux kernel does something that's fundamentally pretty similar; it delays all writes to the disk until a read comes up (because delaying reads would block applications, but delaying writes doesn't). This gives the drive a lot more chance to spin down, because the spindown timer doesn't get reset just so that syslogd can write "--MARK--" to disk. It's slightly dangerous in that your data might take considerably longer to actually reach the disk, but if battery life is really important, it helps :)

Re:Very short lifetime....

[ajlitt]

Good point. Like you mentioned, you are trading some reliability for that power savings. The flash method relies on the fact that IDE (or whatever bus) writes are fairly power-inexpensive, and that the flash at the other end of the bus is nonvolatile and consumes marginally more power than the system RAM but less than the spin-up-and-write-to-disk operation.

On the other hand, modern xDRAM is fairly power efficient at doing refreshes (auto-refresh command vs. old skool mem controller refresh), so I'm surprised that a separate smaller (cellphone type battery) DRAM keepalive battery isn't used in modern PCs. Most laptops will keep the RAM alive in a suspend mode, but this is driven from the main cells. Modern RAID controllers do what I describe here, and can dump unwritten contents of the cache to disk on the next power-up after a power fail. Wouldn't it be nice to have that on laptops or even desktops?

Seems to me that a better idea might be

[ErikTheRed]

Using flash (or, better yet, MRAM - faster, unlimited writes), to hold disk metadate, file and folder allocation information, etc, rather than just as a giant write-back cache.

Support for this would have to be included on a filesystem level, but if this were available I'd imagine the FOSS community would have it testable in a few days, and stable enough for general use in a few weeks.

Re:Who is driving?

[Anonymous Coward]

Montreal?

hahhahah

Re:ho80

[Durinthal]

Apparently the trolls haven't even visited that link recently enough to realize the site is down there as well.

Can we please stop calling everything technology?

[cgenman]

They're using 64 bit technology. The other guys have dual-memory technology. I have technology on my wrist that allows me to tell time.

Can we please stop calling everything technology? At one point the word had meaning, but it's been so over used now that it means nothing. Now it's just a way to make something look more impressive than it actually is, a for-nerds buzzword. "Our emergent 64 bit technology allows for vertical integration along all of your supply-chain specifications." It's a painfully overused buzzword. 64 bit technology. Plastics technology. SUV technology. Technological technology.

Some things still deserve the term. Pretty much anything fusion-related can be given the term fusion technology. But the term technology is being applied to a lot of things that are just design choices. Win 3.1 could have had 64 bit memory addressing, they just didn't because it would have been a huge wasted of prescious resources. Calling it "64 bit technology" is like saying a car has 4-door technology: it's a design choice, not a radical piece of tech.

And these damn kids keep throwing their frisbies on my lawn.

Doesn't solve the problem

[holyshitholyshit]

The problem of keeping the hard drive spun down is solved by having dynamic RAM in the drive. Having flash in the drive only saves a few seconds flushing the RAM before system power down. But flash is much slower in writing than is RAM.

It seems to me this is just an excuse to boost up the prices of hard drives.

Re:Doesn't solve the problem

[mp3phish]

Obviously didn't read anything about this tech...

This isn't anything like dynamic ram in the drive. It is a static buffer which does not require being written to the platter. When the system crashes or power fails, you don't lose your data while its on flash memory. In a memory buffer you do..

This is NOT a cache. It isn't used as a faster buffer between your system and the platter. It is NOT used to improve performance. It's ONLY PURPOSE is to be written to while the disk is spun down. This prevents the disk from having to wake up while you are browsing the web. It prevents the disk from having to spin up when you save a 2K word file. It prevents the disk from spinning up just because you sent an email. No amount of volatile memory is going to prevent the disk from spinning up when it needs being written to.

Re:Doesn't solve the problem

[holyshitholyshit]

You don't seem to realize that the purpose of a cache is precisely what you are claiming is so novel about flash: it exists to permit your computer to avoid spinning up the hard drive. The drive cache is not only for caching data that was read.

Re:Doesn't solve the problem

[petermgreen]

the thing with write caching is people expect when they hit save that thier document hits nonvoltile storage pretty damn quick

this means that if anything is saved on a system with traditional cache systems (without elaborate systems like flash or battery backed ram) the system has to spin up the drive soon after you hit save

on the other hand if the data can be stored in a non-volatile solid state manner by the drive there is no need to spin up the drive at all for small writes.

Re:Doesn't solve the problem

[mp3phish]

You don't seem to realize that the purpose of a cache is precisely what you are claiming is so novel about flash: it exists to permit your computer to avoid spinning up the hard drive

You don't seem to realize that you are completely wrong on this point. Drive cache MUST spin up the hard drive any time it has something to write to it. The reason the cache exists is to the OS can write and forget about it quickly. The cache then writes it to the drive when it has a chance. This does not mean the drive cache will store the data indefinately until the drive can justify spinning itself up. It means the drive must wake up immediately and take its data before the power fails or a system crashes.

If the cache behaved like you say, saving a text file wouldn't actually save it. Changing a system setting wouldn't actually change it. and downloading a PDF file wouldn't mean that you can retrieve it later.. even if you used the save button.

Again, this flash memory is SLOWER than cache. It isn't used as a cache. It is a temperary static storage space that allows you to write data PERMANENTLY without spinning up the disk. No amount of cache will allow you to do this (unless you have a battery backed up cache like they have in high end storage systems.. event hen you are reliant on the battery life of the backup).

Not Nuff

[eSims]

"Let's suppose you had 10GB of primary memory--probably everything that you do could fit in memory," Allchin said.

Yeah, everything except Longhorn :-p

Samsung will probably mess this up, too

[Anonymous Coward]

their "oh-so silent" S-ATA drives are crooked, stay away from them if you want to keep your data longer than a few months...

I bet this flash/hdd will be a flop... ;)

RAM drive

[Masq666]

Wouldn't it be better to use RAM instead of Flash memory on the HDD's, of course RAM can't store something permanent but it's faster. Well most/all HDD's already have some MB's of Cache on them, but if they had more you could possibly use a RAM drive for swap space and leaving your main RAM untouched, RAM is as far as i know faster than HDD's so it would be a speed gain it the OS supported this and loaded itself onto this RAM drive. This may be a bad idea so correct me if I'm all wrong here.

Re:RAM drive

[mp3phish]

This may be a bad idea so correct me if I'm all wrong here.

You have valid concerns, but you have to watch out and not go with what everyone else on slashdot is saying...

This flash space on the drive has nothing to do with improving performance, etc etc... It's ONLY concern is to help prevent the disk from having to spin up when you need to write to it.

Imagine using MS word while you have autosave on. So every 2 minutes your 2K word file gets written out to disk. Well, you are on battery.. your power management settings says to spin down after 2 minutes of inactivity. But if you have word on autosave, then you have to spin back up.. every time.. Extra power requirements. Extra wear and tear on your drive.. With this flash memory you can write out all those small files to the drive without spinning it up. And it will be permanently stored on there. You cannot do this with volatile memory technology.

After you do enough 1K writes to the drive, all of your 128MB of flash will be taken up... Then you will have to bite the bullet and spin up the disk... But the whole point is that MOST users MOST of the time aren't doing that much disk activity.. and so they can go for hours without having to spin up the disk.. thus saving a good hour or 2 of your battery life.

longhorn 64bit memory support

[yakumo.unr]

TFA reads "With its 64-bit technology, Longhorn can support up to 128GB of main memory"

that should be 16TB, really quite a lot more...

source : http://support.microsoft.com/default.aspx?scid=kb; EN-US;q294418

Re:longhorn 64bit memory support

[Anonymous Coward]

They probably had an error in their code and didn't bother to fix it. Microsoft's original AVI spec? Could support files up to 4GB? Limited to .. uh .. about 1GB? Either that or the guy is just a marketing drone in disguise.

Hybrid hard drive

[noidentity]

We should all support them in their effort to reduce our dependence on oil!

Write data to main memory?

[Anonymous Coward]

Sounds like a great idea. Now when Windows crashes, I'll lose my recently-written data, also. Am I missing something here?

Don't try to patent it...

[dolphinling]

I like the idea. Just as long as they don't try to patent it, because I already came up with it on my own.

Re:Moral of the Story

[tomstdenis]

Apple has had it share of horror stories too.

The problem isn't so much as Compaq as just plan lazyness and greed [which the industry as a whole shares].

Generally I don't see laptops as such a great buy unless you can try it before paying... specially with all the wintel hardware out there...

Tom

not as good as it sounds

[distantbody]

it would improve write speeds and energy consuption sure, but this idea falls flat on its face when you consider that the other half of a hard drives job, that is, reading data, isnt improved at all!

Amazing MS Quote

[beswicks]

Microsoft Windows chief Jim Allchin said in a recent interview. "Let's suppose you had 10GB of primary memory--probably everything that you do could fit in memory," Allchin said.

Probabaly!!! What the hell is the average person going to do on a laptop that need 10GB of memory? Nice to know that the MS Windows chief is aiming for lofty goals in efficency.

What sort of pig is longhorn going to be?

[child_of_mercy]

10GB???

Just about everything anyone on a sane OS wants to do should fit under 1GB.

Re:What sort of pig is longhorn going to be?

[amliebsch]

Just about everything anyone on a sane OS wants to do should fit under 1GB.

Are you saying, that 1 GB should be enough for anybody?

Re:What sort of pig is longhorn going to be?

[child_of_mercy]

Well I can run Knoppix fine on 512.

People doing media editting would have a different view of course.

And I'm not talking about financial insitutions backend machines either.

But those extreme uses aren't going to be happy with a one-size-fits-all OS solution anyway.

Flash with Drive on ATA or SATA Controller.

[Embedded]

The point is an old one. It's the buffer silly. iPod's do exactly this for power and jog proofing reasons for 25 minuites. I am *surprised* to not see an integrated drive from Hitachi Travelstar not chucking the ram buffer away and putting in a 256MByte 1 chip flashdrive available today for PCMCIA flashdrive (and today the sweet spot *is* 256MBYTE). HP should love this for PDA's and Laptops. Sure beats Centrino for power savings. Using the internal HD controller for this manipulation resolves dependancy on a Journaling File System. Or the old battery backed up ram.

Re:Flash with Drive on ATA or SATA Controller.

[gru3hunt3r]

No that doesn't make any sense.
Hitatchi's won't add a 256mb flash disk because they can't put that on the spec's.

For example for the same cost which one do you buy:
1. x brand laptop with 1gb flash memory
2. hitachi brand laptop with 768mb of memory and 256mb of flash.

99.9999% of consumers aren't smart enough to make the comarison, even if Hitatchi claims their battery lasts longer -- it doesn't matter because it's impossible to prove, Brand X will claim the same battery life (which is difficult to measure by any standard). And of course Brand X has more memory so personally i'd buy that one.

You and I both know the hitatchi would end up needing to the same amount of physical ram to compete -- and therefore will end up costing $35-$50 more simply because it has the built in flash drive, and that increase in price would ultimately make it less attractive to the consumer. Why would Hitatchi do that? Educating consumers is hard and expensive.

Also the ram buffer can be put to better use when the laptop is plugged in -- even if it does consume more power and heat when it's not plugged in.

NOTE: I've spoken with laptop designers and suggested they add a flash disk, but not for the reason you suggest. Laptop with flash disks could pop in and out of hibernate in like 2 or 3 seconds instead of the normal 30 seconds to minute of current systems. For certain applications this would make them much more attractive. (I would personally give a lot of purchasing preference to a laptop that could do a full boot in 5 seconds -- i'd easily spend an extra $100 for this feature)

I remember when I got a computer that had a 32mb of memory, I copied my Windows 3.0 onto a ram disk in xpanded memory and it started and initialized in about 4 seconds. Wowee! Pretty much everything sense then has been a disappointment.

Using hardware to solve a software problem

[leehwtsohg]

You don't need flash memory or a battery backed RAM on a laptop to reduce power usage. A laptop has a battery. You know exactly when that battery is going to die. A stable operating system would just use the RAM of the computer as a battery powered backup, and write back when power gets low, if neccessary. A laptop is not like a desktop where someone could trip over the powercord, or test the building's main power switch. You have power for hours. (and some laptops even have enough power for 1 minute battery swaps)

But, there is another reason for this: badly written software. It is really hard to get windows not to spin up the hard disk. I have never managed on my windows XP tablet. Even when you close all application, and almost all services, windows keeps doing things like creating the file and directories "C:\Documents and Settings\All Users\Application Data\Microsoft\Provisioning\store.xml".
Actually, linux suffers from similar problems. Even linux is hard to get not to sping up. Some programs are spin-unfriendly. (like KDE...) But, in linux you have more control. And, you have hacks like noflushd. After much work and hours of waiting for the HD to spin down, and stay that way for long enough, I managed sometimes to get a laptop to spin down and stay that way for a long time.

But, as I said before - this is a software problem, and can have a software solution. Everything you need is already in the computer - RAM battery backup, CPU.

Re:Using hardware to solve a software problem

[Anonymous Coward]

Exactly! I would guess this has a lot more to do with M$ Windows inadequacies than anything else. Maybe they realized that their software spends 95% of its time asking the files on the HD the same questions that were first written during install? Instead of re-writing all of that code, they add a flash card that they query a billion times. Hell, maybe they just write the registry there? God knows Windows reads that everytime it runs an app and only updates it when a new app is installed. It will probably also keep default user preferences for software as well, basically anything you might write to the disk once a month.

We also haven't talked about software license keys or DRM. All of that info would go here, too.

Thought about in this way, it makes a lot more sense.

John

You are not get it...

[tcdk]

If you had a large RAM lazey-write cache you would loose everything come that big crash.

With a flash based cache on the drive, the drive would just keep on writing the next time it gets power.

The low access times of this, would actually mean that you could do some funky stuff (like moving data around in a file) really fast with out fear of loosing anything.

For databases think how this would impact transaction based things...

Re:Can we please stop calling everything technolog

[Anonymous Coward]

"Technology" is not a buzzword. It is simply a word meaning roughly "application of tools". In other words, 64-bits is another tool to be applied, thus if Windows applies it, it is using 64-bit technology.

Don't forget, hammers and wheels are technologies too, just commonplace ones.

dom

A way to use flash at the OS level?

[Anonymous Coward]

This was a little interesting to me to see because I have been wondering if there was a way to exploit the properties of flash memory at the OS level, but I guess that might be coming in the later OS.

For example, you don't need to be tampering with the program executables too much most of the time, usually you only write the files once and read from then on. This fits perfectly with flash memory which has limited writes but unlimited reads.

Given that, my idea was to buy a flash card of large enough capacity to store the OS and load the entire OS (and possibly program executables) on the flash drive instead of a hard disk. The only time the hard drive would be necessary is when dealing with data that needs to be modified often and the only files that are modified often are user data, configuration files, and virtual memory. Unfortunately, many pieces of software do not allow you to seperate static data (executables) from dynamic data (configuration, settings), so this wouldn't work very well without some sort of OS and program support.

Samsung becoming the new Sony

[glitch23]

There is an article in the new Wired issue about Samsung becoming the new Sony and becoming the leader in LCD manufacturing. Very interesting.

Why use flash? Why not battery backed RAM?

[TheLink]

Why not just use battery backed RAM? It doesn't have to be the super-fast power hungry desktop RAM. 100+MB/sec would be more than good enough.

Stick in a few of those low self-discharge "watch" batteries that last for years.

If the batteries run out, just behave like a normal harddrive.

Then you'll have a reasonably fast nonvolatile cache, and the drive can take its time to figure out what is the best way to write out all the data.

Re:Why use flash? Why not battery backed RAM?

[mp3phish]

Why not just use battery backed RAM?

High end storage systems do this already.. It's a good idea... But they do it for another reason: to gain extra performance while under heavy load, and still prevent data loss on a system crash or power failure.

I believe (but may be wrong) that using this cheap flash memory in here is lower maintenance, cheaper to produce, less complex of a design, and more reliable than the battery backed up system. And since this isn't ment to improve system speed, but to prevent the drive from having to spin up while in a low power state, then it is a better solution to this particular problem.

Though you are right, battery backed up ram would probably do just as good of a job.. just more expensive.

Re:Why use flash? Why not battery backed RAM?

[TheLink]

Yah, maybe they should build them into high-end drives or something :)..

Anyway, in a few years probably MRAM would be a better choice. Spinning disks may be cheaper for some time, but the MRAM would be good for a nonvolatile buffer.

Re:Can we please stop calling everything technolog

[Anonymous Coward]

langauge, writing, speeh, symbolic thought, musical notation.
musical scales, intruments . etc

however... i agree with its overe use in the 'geek-sphere' - and denigrates the use of the term above.- we arent impressed more or less. we dont like market speak short hand do we?
so hey i agree lets not put up with it.

Couldn't this be implement it as a Linux driver?

[Anonymous Coward]

Couldn't this be implemented quite simply with a userland driver on Linux.

Maybe a variation of the Knoppix CD-ROM driver can be hacked up over a weekend to do the same as proof-of-concept (I can't do this but maybe a reader can). Or maybe a reiser4 module can keep all FS transactions on the Flash cache until it hits a certain boundary after which the HD is updated using existing code.

The cost would then be simply that of a USB Flash plug (I guess most of us have one already anyhow). Using an external plug it would be also upgradable ... plus as userland driver one has the benefit that the algorithm that tries to optimize the write-to-Flash-spin-up-HD-flush-multi-cache cycle can be made so much smarter than anything they can put implement on the harddisk itself.

This could see this be universally useful for desktop systems or systems used as TVs to keep them cool and quiet.

Uh, what about READING?

[Kris_J]

I have a 5Gig Seagate USB drive. While it might be months before I've written 128MB of changes, when I'm using it I'm randomly referencing over 4Gig of data files. How would a little flash cache save it from spinning all the time?

OS Support for Flash Caches vs. Disk-controller

[billstewart]

If you build the flash cache into the disk controller, it has a very limited functionality and can therefore mostly hide it from the operating system except for maybe providing some energy-star hooks. It's nowhere near as useful for some applications as a general-purpose flash cache would be (e.g. database journaling), but it's easy to add, and it gives the disk drive manufacturer something to sell that might have a bit of profit margin hidden in it.

On the other hand, if you put the flash somewhere that the operating system can use it directly (whether that's on a disk bus or a memory bus or somewhere else), you get a lot more flexibility - at the cost of needing to do operating system development and/or application development to support it. Application development may be easier, e.g. tell the database that /dev/hdc1 is flash and it'll know how to use it, or write a file system driver that does something reasonable with it, as opposed to messing with the entire Virtual Memory, caching, and memory allocation strategies.

OS support is a big problem for Microsoft - they're huge and slow, Longhorn is late, and when I last kept track (a decade or so ago) they really weren't very good at managing caching (and given how XP and Win2003 behave when Mozilla-with-50-tabs gets too big, it still feels like they're not very good at it.) Linux or *BSD or MacOS could integrate it much better, because Unix platforms have a head start on them, but it would still need to be a major change.

OS/controller support for Permanent vs. Cache?

[billstewart]

I didn't get from the article that this was for permanent storage - how is the disk controller going to know what to store in flash and what to store on disk? If it's a cache thing, then the controller can just cache each write and the OS doesn't need to know when it actually got written to the disk - it just gets the "write succeeded" ack back faster and reads to cached material are really speedy. But if it's a permanent-storage thing, that implies that what it really looks like is that the first 128MB or whatever of /dev/hda is really in flash, so the OS needs to know to store the interesting stuff in that section of the disk. That's already a problem, and once you start using applications, it's much harder to define what the "interesting" stuff is that ought to be in that fast-storage space so that you can avoid having to hit the disk for the application that's currently running.

Re:OS/controller support for Permanent vs. Cache?

[mp3phish]

This flash isn't a "fast storage" space. It exsits solely to be written to while the disk is spun down. It has no other purpose.

Trying to say it will improve performance by reading faster than disk is preposterous. This type of flash memory is not only slower, but has a limited write lifetime. Thus it could not be used effectively as a cache.

The permanent storage I speak of is when you write to it while the disk is down. If the power fails or if the system crashes, it is permanently on the disk and can be retrieved later. With a cache this cannot be done. The OS doesn't have to make difficult decisions on what to store on this 128MB flash chip. It will only store things which need writing while the disk is spun down.

Laptop Flash vs. Desktop RAM caching

[billstewart]

Laptops and desktops/servers are really much different problems. A laptop has limits to power, weight, size, and heat load, so you wouldn't do a battery-backed-RAM disk cache system for use in unplugged mode - you'd want to do something with flash, and your goal is minimizing power and disk usage, not maximizing speed.

Desktops and servers are a much different game - your goal is maximizing speed, size&weight aren't a problem, and you have lots of electricity available to power everything except when Something Bad happens. So you can use standard RAM (which can use a lot of power) and build in batteries or use a UPS, and while you could integrate it into your disk array appliance, you could also just use it as a separate drive and tell your application or your OS to use it. ($50K probably does get you a really cool system, but a $200 PC, $250 worth of RAM, and a $50 UPS can get you a 2GB system with Firewire/USB/GigE interface for ~$500 if nobody wants to make something better and integrated.)

The old Legato Prestoserve from ~1990 had about 1MB of battery-backed RAM, and was designed as an accelerator card for Sun's NFS Networked File System and also for database applications - it was big enough to make an enormous performance difference, because it eliminated disk seek and rotational delays when storing journals and committing writes, and only needed to handle ~100 milliseconds worth of data. On modern disk arrays, obviously the transfer speeds are much higher, but the seek and rotational times are shorter, and the price of RAM is much lower, so might as well pack in a couple of GB.

Did they provide more detail about how it works?

[billstewart]

The press release wasn't very deep or very long - did the trade show folks provide more information about how it worked? 40 years is obviously not the working time for something that caches every disk write, so they're doing something to decide what to cache and what not to cache, which seems to be driving much of the Slashdot discussion.

I bet you do frequent backups, too!

[billstewart]

ok, ok, I shouldn't feed the trolls, but I recently had my laptop disk die from heat death and the loaner PC get stolen while it was in the shop :-)

If you really have a good backup system, with everything updated to a server or desktop at least daily and preferably constantly when you're online, and some way to restore a machine image rapidly, then you *might* not lose several days of work reinstalling everything when your disk gets trashed *after* the corporate IT droids back in New Jersey have mailed your ostensibly repaired machine back to you.

Alas, typical support isn't that good, so you're going to lose more than just the morning's new typing. And just because you're getting your salary, that doesn't mean that the people you're supporting aren't handing you work to do while you're wasting your time with the bloody broken PC.

MOD PARENT UP PLEASE ( + Comment )

[billstewart]

It's definitely an Insightful article, but I've posted several comments on this thread so I can't also moderate it.

On the other hand, laptops usually *don't* really know how much power they have left - battery behaviour is much less linear than you'd expect, and my experience with the battery gauges and automatic-save systems on the last half dozen laptops I've used is that they're pretty accurate when the battery is new, and increasingly unreliable as the battery gets older, so it fails to do the friendly gentle save mode while it still can, as opposed to the "flush a bit of cache and then shutdown" mode, which loses all those open windows you had.

Windows NT 3.51 was the probably the worst experience I've had with this, since MS was pretending that NT was a "Server Operating System" and not intended for desktops, so it didn't have power management support. I was commuting by train, with a laptop battery life approximately equal to the train trip if I caught the express. The hardware was smart enough to detect a low battery and send an interrupt to the operating system to tell it to save everything, but the OS wsn't smart enough to respond properly so it would blue-screen. Sometimes I'd get ten seconds of about-to-shutdown beeps which I could use to save my files, and sometimes I wouldn't.

That's not typical user behavior

[billstewart]

While there's a lot of material you're occasionally using, for the most part most people's usage doesn't behave like that unless they're watching movies or something. A few megabytes of real text wrapped up in some bloated Office format, plus some programs that get cached in RAM (MS may not be the world's experts at caching, but they're good enough to do that much), plus occasional bursts of swapping which may be annoying but aren't randomly spread out on disk.

Also, if you're using Microsoft Outlook for your email, you might be writing out large amounts of changes, depending on how they manage space within the .pst file (assuming you're using one, which is a good assumption for laptop users.) Unfortunately, the format doesn't appear to be documented in a way that anybody but Microsoft employees typically has access to.

Re:That's not typical user behavior

[Kris_J]

Might be more typical than you think; I was describing what happens when I play World of Warcraft. (The drive is quite warm by the end of an hour session.)

Throughput vs. Latency,Spin,Startup,Battery

[billstewart]

I'm not sure if you're saying flash is slower than disk, RAM, or what. Reading from flash may be slower than reading from DRAM, but it's reasonably fast. Write speeds are also much better than they used to be (don't get distracted by the slow USB1.1 bus your older flash-RAM stick has. A quick Google search found flash sticks with write speeds of 5-13 MB/sec.) and people have been playing interesting games with RAID which could easily be built into an on-board customized controller like this.

But the big difference is that disk drives are rotating hardware, so they consume power keeping them rotating, take time to stop and start rotating, and take time to rotate to the right location when you want to read or write something. Flash memory doesn't have as fast a megabytes/second writing throughput as a rotating disk drive, but there's no latency, and the amount of time it takes to write to flash is fairly reasonable compared to the time it takes to start up a spun-down disk, wait for it to stabilize, wait for it to rotate to the right location, and then blast data onto it. And it lets you leave the drive spun down unless you've accumulated 128MB-1GB of stuff to write, or unless you need to read something you don't have cached in memory somewhere, so you save a lot of battery.

No need for it to be close to the processor as opposed to close to the disk drive or somewhere in between - that's an architectural question driven mainly by software capabilities, and flash is slow enough that it can live out on a disk bus if that's the right choice, unlike video RAM or something. The prices *would* be higher, as you suggest (which I'm sure the disk drive companies don't mind), but it looks like this is being driven by Microsoft rather than the disk companies.

Knoppix etc in Flash instead of CDROM works fine

[billstewart]

USB-based Knoppix FAQ. If you can run Knoppix, Damn Small Linux, etc. from a CDROM, you should be able to run them fine from a USB Flash stick as long as your PC and USB stick let you boot it (some USB sticks aren't bootable, some look like CDROMs, some don't.) You need to do slightly different setup so it'll find everything, handle partitions appropriately, etc., and it'll use more RAM than necessary if you don't tune it a bit. (If you're a Windows user, the official answer about whether you can do this with BartPE is "maybe" and they expect it'll get better in the future. I don't know if Macintoshes can do anything similar.)

For software that's grumpy about where things are loaded, the easier solution is to run it from disk rather than flash - it'll still cache in RAM, which is mostly just a startup-speed issue, and you'll have most of the OS and applications in flash so you'll get most of the speedup you want. There are fancier solutions like translucent file systems of various sorts that let you mount dynamic stuff on top of the read-only stuff, or you can play with partitions some more (e.g. use the flash as native storage rather than compressed cloop stuff.)