Samsung Ships Hybrid Hard Drives

writertype writes "ExtremeTech reports that Samsung has become the first company to begin shipping hybrid hard drives as discussed last fall on Slashdot. (Some photos here.) Unfortunately, there's no word yet (beyond 'soon') on when retail shipments will begin, or when (or if) 3.5-inch models will be available. Note that these hybrid drives are different than the ReadyBoost USB flash drives optimized for Vista; hybrid drives contain a smaller amount of flash, and work as a write cache for your notebook drive, extending battery life."

Comments

Linux

[Constantine XVI]

But, really, can they run Linux? Are the drives supported in the kernel?

Re:Linux

[LWATCDR]

They probaby use a SATA interface so no driver other than one for you SATA controler will be needed.

Re:Linux

[TheThiefMaster]

Who cares? I'll take a wild guess and assume people who run linux might care. I'll take an even wilder guess and assume that the poster who asked the question cares.

Re:Linux

[morgan_greywolf]

If they use a standard interface like SATA, IDE, or even USB storage, then no special driver is required in the kernel.

Re:Linux

[compro01]

to run the drive perhaps, but what about to use the caching? is the "write to buffer till full, then dump to disk" thing handled completely within the drive firmware itself or does it depend on OS-side drivers? TFAs are kinda sparse in that info.

Re:Linux

[jimstapleton]

Thinking of that, what will it do to data integrity on crashes and power failures?

Re:Linux

[jimstapleton]

diregard... I was forgetting that flash is non-volatile - It'll still be there for recovery on reboot.

Re:Linux

[Anonymous Coward]

They probaby use a SATA interface so no driver other than one for you SATA controler will be needed.

Do you know this or are you just pulling this out of your ass?

Re:Linux

[Jeff DeMaagd]

A driver is probably needed to handle the hybrid part - to know what to do with the special features that are new to consumer drives. I think the OS has to decide what to put on the flash cache, I don't think that the drive can realistically be expected to do that on its own. With a current generic driver, I don't expect that there would be any benefit to using this type of drive.

Re:Linux

[UnanimousCoward]

Old news. Move on. These companies and others have been offering them for years. And they run Linux...

Re:Linux

[AlecC]

Almost certainly completely hidden by the driver interface. Firstly because it would be easy to do, secondly because I doubt that (e.g.) the SATA interface has commands to handle this new technology invented about eight years after it was designed, and thirdly because keeping it built-in vastly simplifies the power fail handling.

Re:Linux

[rblancarte]

Not at all. Hardware to do all that you are talking about is probably on the Drive side of the SATA port. It would be transparent to any host system because of the SATA interface. All that it would care is that it sees a SATA drive, and it appears to be really fast!

RonB

Re:Linux

[LWATCDR]

Many drives already have a read cache in ram. The drive electronics figure out what to cache and what not to. Sata, EIDE, and SCSI drives all have some kind of microcontroler that may handle the cache. Since the samsung site only seems to work with IE and I only have Firefox and Opera on my Linux box I have to guess.

Re:Linux

[CastrTroy]

Well then how come MS is touting this as a new technology in Vista? If this was really something that the drive handled then it would work in any OS, including Vista, XP, Linux, 98, etc.

Re:Linux

[captain_craptacular]

If you read the blurb, or the article, this is not the same thing as ReadyBurst TM which MS touts as a new feature in Vista. ReadyBurst TM, allows you to plug in a flash drive and use it as a sort of replacement for part of the disk. ReadyBurst TM allows you to use up to 2GB of flash. This technology puts the flash right on the drive, uses a much smaller amount of flash (128-256MB is optimal), and is more about power management than speed increases. Think of it as a relatively large, non-volatile cache on the HDD. The drive can use it as a buffer until it's full then write the data in one concentrated burst. This way, the power hungry spinning drive only has to spin occasionally.

Re:Linux

[profplump]

The SATA interface most certainly has commands to handle this new technology -- you can send arbitrary commands over SATA, just like you can over SCSI. It's a generic data interface not a block-layer device controller.

You'd just assign the controller another LUN and document the commands it accepts. You could then make the flash disk part of the address space of the primary disk or you could assign each their own LUN for use as two separate disks, with the third "control" LUN accepting commands to copy between the first two out-of-band. You could even setup a system where "dumb" controllers could use LUN 0 and treat the disk as a normal hard drive, with either automatic or no use of the flash portion, and expose the same physical disks separately on other LUNs for advanced use by "smart" controllers. Take a look at any tape library or even an iPod for a similar example.

Re:Linux

[JoGlo]

If you read the blurb, or the article, this is not the same thing as ReadyBurst TM which MS touts as a new feature in Vista. ReadyBurst TM, allows you to plug in a flash drive and use it as a sort of replacement for part of the disk. ReadyBurst TM allows you to use up to 2GB of flash.

Close, but no ceegar, I'm afraid!

ReadyBoost doesn't replace anything, really. What it does do is provide a very fast cache for files, that are also saved to disk, but can much more quickly be accessed from the flash drive (which can be a memory stick, SD card, CF card, or whatever, provided it is overall fast enough for Vista to use (it does a complete R/W check to make sure that the device is fast enough across the whole device.

Ideally, for those with smaller amounts of memory, a ratio of 1:1, flash : memory, but with larger amounts of memory, 2.5 : 1 is the recommended amount of flash drive to use, up to a maximum of 4gb, which is the limit for FAT32 apparently.

Re:Linux

[CastrTroy]

The limit that MS puts on FAT32 is 32 GB for a drive. The limit is actually much higher, and if you format a drive with Linux or many other disk utilities you can get really big drive. Up to 2 TiB, according to Wikipedia. The max file size is 4GB, but I don't see how this would really affect the maximum of this application, as they could just use multiple files to store the data.

Re:Linux

[morgan_greywolf]

Not quite.

4GB is the max size for a FAT16 drive. 2GB is the max file size for FAT32. (Trust me, I know from practical experience in trying to move large Oracle dumps via a FAT32-formatted USB drive, and I was there for the 4 GB FAT16 limit.)

Re:Linux

[CastrTroy]

I put 4GB files on my FAT32 formatted hard drive all the time without any problems.

well

[mastershake_phd]

Isnt flash only good for ~30,000 writes? If the flash breaks, can you still use the drive? And most importantly how much does it cost? I think the spinning magnetic disc is still king for a while to come, unfortunately.

Re:well

[bendodge]

A new kind of flash was developed last year that had much faster read/write (closer to RAM) and didn't deteriorate. I suspect that kind is what these will use. (Unfortunately I don't remember the name...)

Re:well

[TheThiefMaster]

Wikipedia says that NOR flash is good for "10,000 to 1,000,000 erase cycles" and NAND flash has "ten times the endurance". Lets hope they've used the good stuff.

Re:well

[Anonymous Coward]

BSFlash

Re:well

[yog]

It's more like 1,000,000 writes, but your point is taken. Perhaps the driver takes this into account--store many small and frequent temporary files such as browser cache files into RAM rather than flash, then dump them all to flash or disk rarely, but this implies a lot of intelligence on the part of the driver.

According to PC Mag link from the article, only Vista has the correct driver to use this drive.

It sounds like a nice innovation. Now to get from hybrid drives to biofuel laptops that run 8 hours on a thimble of ethanol ;)

Re:well

[SScorpio]

As someone else has already stated, yes flash is still limited but not as much as it used to be. These hard drives are aimed at laptops and I believe Vista requires them to be considered as "Designed for Vista" rather than Vista Ready.

The point of the flash is to provide a nonvolatile write cache which will then spin up the drive to write a queued data after the cache is filled. This is supposed to have a significant effect on the battery life of laptops.

Re:well

[NMerriam]

Isnt flash only good for ~30,000 writes?

The have limited cycles per sector, but the drives automagically allocate writes over the least-used sectors. In practice, a modern flash drive should have at least the same lifespan as a spinning disk if not longer.

Re:well

[TheThiefMaster]

I'd expect the drive to have a normal (2-32MB) cache as well, which it will use to buffer the data before writing it to the flash, especially as flash can only write data in blocks.

I'd also hope that in heavy usage it disables writing to the flash and behaves like a normal disk to avoid wearing the flash out.

Re:well

[necro81]

I suspect that the intelligence built into the drive has the capability of detecting flash sectors that have gone bad, much like an ordinary hard drive can detect bad magnetic sectors. So, I think that over time one will see that the flash's capacity decreases, but is mostly still available during the life of the drive.

Re:well

[Jaseoldboss]

PRAM has the properties you describe. Although it isn't a type of Flash memory so I doubt it's the one present in hybrid drives.

Re:well

[justthinkit]

Who would be reassured by the following:

The average human is good for 10,000 to 1,000,000 hours.

Re:well

[Nasarius]

Linux must get full support for NTFS.

*tap tap* ntfs-3g -- I'm using it now, and it's performing nicely even under pretty heavy BitTorrent load. ntfs.fsck still needs to be written, but the situation is now vastly better than it was less than a year ago.

Re:well

[Anonymous Coward]

MRAM probably. It's similar to flash in that its non-volatile, but it's not the same tech. You can buy it, today, from Freescale Semiconductor, except it's very expensive.

Re:well

[EnderWiggnz]

More than a typical spinning HD, actually.

MTBF in NAND flash is between 1M hrs and 3M hrs. They don't even use write cycles any more.

Re:well

[TheThiefMaster]

Who would be reassured by the following:

The average human is good for 10,000 to 1,000,000 hours. That's surprisingly accurate:
10,000 hours = Roughly 1 year, 50 days
1,000,000 hours = Roughly 114 years
Most people do die in that timespan, even if it is a little broad.

Anyway, back to flash: Those numbers aren't from the same variety of flash, they might be using one that averages say 800,000 erase/write cycles, with 99.999% of devices being within 50,000 of the average. I certainly wouldn't mind knowing how long I was going to live that precisely, and I definitely wouldn't mind living 800,000 hours (I'd be 91!).

Re:well

[maxume]

Newer flash is supposedly closer to 100,000. Also, 128*100,000 is larger than 8*30,000 in a useful way. (for the lazy, by a factor of ~50)

Re:well

[SkewlD00d]

Hi,

Flash is typically *rated* for 10^5 writes.

I worked at trimble navigation, radio group in sunnyvale, ca in the summer of 2000. One of my projects was stressing flash eeprom in the embedded systems we were developing, using rapid thermal cycling, and finding ways to exceed and recover flash beyond manufacturer's rated duty-cycle spec. Yes, we all know this is similar to MTBF calcs and not the same as real world failure modes (*cough* google's hard drive paper). The funny thing was, flash rated at 10^5 writes, even after 10^6+ writes it simply wouldnt fail. What I read is that the failure modes were generally single bit column errors (multiple bytes with the same error), or sector stuck errors. The way to get longer useful life is simply use bigger flash and 2D ECC.

Regards

Re:well

[charlieman]

Intelligence? do we have to welcome our new bad flash sector detecting intelligent overlords now?

Re:well

[Joe The Dragon]

Well if you just use for OS boot code / system files then it does not need a lot writes

Re:well

[drinkypoo]

ntfs.fsck still needs to be written, but the situation is now vastly better than it was less than a year ago.

Amen! I have ntfs-3g on my Ubuntu (Edgy) partition. So long as I do a safe shutdown, and the filesystem is marked clean, everything works wonderfully and very quickly (not that I had serious speed problems with captive-ntfs, but I seldom deal with very large files.)

It's quite amusing that Linux is the only OS that can natively (as in, as a filesystem, not just in some ftp-like application) handle basically every major filesystem in existence today, what with the addition of NTFS support.

Linux is the only convenient way for me to transfer files from a HFS+ volume to a NTFS volume or vice versa. You can do it on Windows by using macdrive, but that is like using winzip or something. And it's damned slow. You can't do it on macos AFAIK, at least I haven't seen working NTFS R/W on macos yet.

And of course linux also supports a shitload of BSD formats, XFS, JFS, ZFS...

Re:well

[drinkypoo]

So, I think that over time one will see that the flash's capacity decreases, but is mostly still available during the life of the drive.

It's also possible that they put some extra flash on there with some backup blocks, just as hard drive capacity is actually greater than what is reported, but some of that space is saved over for bad block relocation (in addition to simply being able to lock out bad blocks, which is what happens when you run out of relocation blocks.)

Re:well

[kronchev]

My god, you fool. It's just like gasoline-electric cars, there are no real pure-electric cars out at the moment because they have too many flaws, but hybrids are on the rise because it has a little of the electric advantage with the proven performance of gasoline.

Just...think before you post.

Re:well

[tlhIngan]

Wikipedia says that NOR flash is good for "10,000 to 1,000,000 erase cycles" and NAND flash has "ten times the endurance". Lets hope they've used the good stuff.

NAND and NOR flash are completely different types of flash chips.

NOR flash is good for holding code - it's basically nonvolatile RAM. You can execute code straight out of NOR flash easily by hooking it up to a memory bus.

NAND flash is good for holding bulk data. It's interface is strictly I/O based (like a hard drive) - you cannot directly execute code from NAND flash without copying it to RAM first. Some NAND-based devices have fancy tricks (Like samsung's ONENAND and M-System's DiskOnChip) where they put in some SRAM so you can execute, but they basically have to copy it from the array into the SRAM. (NAND flash also has stuff like "bit flips" where read data does not exactly match written data - and reading data can change it, but this is compensated for by using ECC codes in the "spare area").

All NAND-flash handling code has to handle bad blocks as a typical chip can have up to 2% bad from the factory.

The reason we use NAND flash is because it's extremely dense. While flash gets increasingly expensive as you go larger (32-64MiB is the "sweet spot" in price/storage for NOR flash), NAND flash achieves really dense storage. For the price of a 32MiB NOR flash, you'd get 1GiB NAND flash chip easily. So for things like memory cards and stuff which use I/O interfaces, the flash is exclusively NAND. NOR is used for stuff like BIOS code which doesn't change very often anyhow, and often just enough of it to have code where we can pull out data from cheaper storage devices (NAND flash and hard disk, for example).

So yes, it'll be the "good stuff".

Re:well

[compro01]

all the flash chips we use at my college are rated in write/erase cycles, haven't yet seen one listing MTBF.

Re:well

[Beardo the Bearded]

It depends on the chip.

Your average chip (like the 16F88) has a 100,000 write cycle for its internal Flash. The SPI Flash chip M25P*0 has the same - 1,000,000 write lifetime. (By memory - I could be off by 10x on the `88)

Now, since this has come up before, that doesn't mean that your drive will work perfectly until it hits 1,000,000 writes and then mysteriously stop working with a blinking red LED on the top. What that means is that statistically speaking, there's a good chance that most of your chip will still be writeable up to a million times. Some bits will fail sooner, some will fail later.

If you're storing a lot of photos, it may not matter if a pixel in the middle is black instead of green. If an MP3 has a 1-bit blip, you won't notice. If you're storing a lot of financial data, it certainly will matter if the MSB in the millions byte is a 1 or a 0 or the tracking software on your GPS-enabled VHF transmitter tells you that your stolen truck is at 45' instead of at 55'.

So you can "use" the drive even when the Flash is fried. Depending on the data, that may not be a problem.

Re:well

[Anonymous Coward]

A new kind of flash was developed last year that had much faster read/write (closer to RAM) and didn't deteriorate. I suspect that kind is what these will use. (Unfortunately I don't remember the name...)

The answer is: Phase-change Random Access Memory (PRAM); also know to the marketing types as "Perfect RAM".

http://blogs.zdnet.com/hardware/?p=93

Hope it helps.

Re:well

[TheNetAvenger]

It's more like 1,000,000 writes, but your point is taken. Perhaps the driver takes this into account--store many small and frequent temporary files such as browser cache files into RAM rather than flash, then dump them all to flash or disk rarely, but this implies a lot of intelligence on the part of the driver.


Anyone interested in how this is handled, go look up MS ReadyBoost/ReadyDrive/Superfetch technologies.

This was an early issue with the ReadyBoost technology in using USB Flash drives and MS designed the caching system to not use the same bits all the time, so in theory by using a bit of intelligence and not addressing the ram in a linear fashion, even an ordinary old Flash Drive should outlast a HD or the computer it is being used on to improve performance.

The same holds true for ReadyDrive caching technology.

Go to Wikipedia or even Microsoft's site for more technical information on how this works.

Re:well

[itsdapead]

AFAIK, at least I haven't seen working NTFS R/W on macos yet.

macfuse claims to support ntfs-3g under OSX. Looks like you'll have to compile ntfs-3g from source though & mount from the command line - they've only got binaries and a GUI mounter for SSHFS at the moment

Re:well

[drinkypoo]

A couple months ago when I set up ntfs-3g on Linux my googling around indicated that ntfs-3g on macfuse was pretty sketchtacular. I did forget that it existed at all, though. It may be dazzlingly wonderful by now.

Re:well

[dgatwood]

In practice, a modern flash drive should have at least the same lifespan as a spinning disk if not longer.

Only for lightly used flash disks. In practice, you cannot "automagically" allocate writes over the least-used sectors because modifications to data are not distributed evenly across all of the flash chips. Your filesystem metadata will tend to be clustered in the first flash chip, which will result in much faster wear, as it is the most frequently modified data on the disk. As a result, the life expectancy of a flash disk would likely be much, much lower than that of a hard drive.

Fortunately, this is not about a flash disk. It's about using flash as a nonvolatile cache. If designed correctly, you could distribute caching evenly across all the flash parts, and thus they would not suffer from the same sort of premature breakdown problem. My quick math suggests that even still, a disk with 32MB of flash rated at 1,000,000 write/erase cycles (typical for flash parts) will only last approximately a year if you write one megabyte of data per second. While that is a lot for a home machine, I have a hard drive that has been running 24x7 in one server for over 11 years. That would fail with continuous use of less than 100k per second. That's remarkably close to the amount of paging that the machine in question does, not counting any other I/O. For a home computer in a TiVo-like application, the drive would only last 2-3 years, which is also far less than a hard drive under similar use. I'm sure there are plenty of other situations in which a flash-cached drive would keel over long before the drive mechanism would fail.

Also, IMHO, unless the caching is directly under OS control, it is basically a useless technology. Under OS control, you could keep frequently accessed data in that flash cache permanently, thus speeding up boot times, etc. Under drive control, a block is a block, and keeping use counts would just eat the flash parts faster. Besides, data moves around when files are modified, making block-level access counts much less useful than file-level. Using it as a write-only cache only helps for short burst writes, which RAM can and should be used for just as easily.

All in all, I wouldn't touch a drive with this tech unless I could permanently disable it with a jumper if the flash parts started acting up. Even then, I wouldn't pay a penny more than I'd pay for a drive without this tech. It's a lot of hype for no real purpose. Just my $0.02.

Re:well

[NMerriam]

modifications to data are not distributed evenly across all of the flash chips. Your filesystem metadata will tend to be clustered in the first flash chip, which will result in much faster wear, as it is the most frequently modified data on the disk.

Why aren't modifications to data evenly distributed over the flash? That's much of the advancement in modern flash controllers, distributing those writes. Just because your metadata is initially written in the first chip doesn't mean the updates have to be written to that same chip. The controller can put it on whatever chip it fancies. We're not dealing with spinning disks and moving heads, there's no need to worry about the physical location of a sector to optimize data transfer.

Re:well

[dgatwood]

Hmph. Apparently current controllers do multi-chip wear leveling. I did not know that. Regardless, the failure figures I was fiving were based on a theoretical perfect 100% wear leveling scheme across the entire set of flash parts. Any less-than-perfect organization that results in hot spots on the flash would fail even sooner.

Re:well

[mrchaotica]

That would fail with continuous use of less than 100k per second. That's remarkably close to the amount of paging that the machine in question does, not counting any other I/O.

If your server is paging that much, why haven't you added more RAM?

All in all, I wouldn't touch a drive with this tech unless I could permanently disable it with a jumper if the flash parts started acting up. Even then, I wouldn't pay a penny more than I'd pay for a drive without this tech. It's a lot of hype for no real purpose. Just my $0.02.

You're forgetting the primary purpose of this: increasing buttery life in laptops. Even if it rears out after a few years, it could still be worth it for that!

Re:well

[Achromatic1978]

increasing buttery life in laptops

I can't believe it's not exploded!

Re:well

[mrchaotica]

Damn it! Stupid handwriting recognition -- sometimes it changes words on me without me realizing (my "u" and "a" can look rather similar).

Re:well

[JoGlo]

A new kind of flash was developed last year that had much faster read/write (closer to RAM) and didn't deteriorate. I suspect that kind is what these will use. (Unfortunately I don't remember the name...) If you burrow down to the original blog Q & A, the MS guy says that they have been very careful how they use reads & writes, and they expect memory sticks to last 10 years being used like this, so it appears that dead sticks isn't a problem.

Samsung drive reliability

[UncleTogie]

I've seen more Samsung drives fail at the shop recently than any other. I hope they've got their QA metrics set a bit higher for the new drives....but I'm not holding my breath.

Re:Samsung drive reliability

[Aladrin]

And I've seen more X brand drives. (No, I'm not even going to bother naming them, but it's not Samsung.)

As they say, the plural of 'anecdote' is not 'data'.

Re:Samsung drive reliability

[UncleTogie]

I'll relay here what I see at the shop, and mod me up or down.
I'll not back from my opinion 'cause of some mod-happy clown.

hard drives are going away

[Danathar]

I'm certain that hard drives will slowly go away to be replaced with Flash ram devices. As the price drops it will happen.

Reasons?

1. Hard Drive reliability - See the security now podcast or read google's paper about hard drive reliability. The manufacturers are lying BIG time about how bad it's gotten. And SMART is a steaming pile of nothingness that can and is wildly inaccurate.

2. Latency (not speed) is so much better than hard drives.

3. Power and heat - Flash memory does not generate near as much heat or draw as much power. Plus we can expect densities to get higher so the footprint probably will be smaller than hard drives

We've already seen it in handhelds. It's moving to laptops (Toshiba and Fujitsu already are selling laptops)

If it has a mechanical action to it, it can fail horribly.

just my 2 cents.

Re:hard drives are going away

[fizzyabbo]

If it has a mechanical action to it, it can fail horribly.

A succinct description of my love life =(

Re:hard drives are going away

[Anonymous Coward]

just my 2 cents.

Would it be possible at all to get change?

Re:hard drives are going away

[TheAwfulTruth]

I don't see HD's going anywhere for the forseeable future.

Well yes, IF flash ram can overcome it's shortcomings AND cost which is extreme.

you can get 750 gig of HD for $350, probably less now, how much would that cost in flash?

And unfortuantely flash is about as reliable as HDs right now for long term use. Even though it is not mechanical, it still wears out and is subject to out of box failures. (Memory manufacturing is about as poor as HD manufactuing is these days based on the number opf bad flash mosdules I've run into.)

And... it is so very very slow.

So yes, it woulf be GREAT to get rid of the bulky, loud, power hungry, slow access, mechanical HD of the last century, but... there is really nothing even close on the horizon right now :( Sadly, flash just isn't practical at all in it's current form for anythig OTHER than small devices that only need a small number of gig in a tiny form factor.

Re:hard drives are going away

[tknd]

Hard drives aren't going anywhere. The more likely scenario is that both flash and hard drives will coexist to exploit the benefits of the strengths of each medium. With a hard drive you can get high transfer rate and high storage capacity. With flash you get low latency and low power consumption.

In fact it's already happening. Windows has that readyboost stuff and samsung is developing these drives. All that's really happened is we've added another type of memory to the hierarchy: registers, cache, ram, hard disks and now flash.

Re:hard drives are going away

[owlstead]

"So yes, it woulf be GREAT to get rid of the bulky, loud, power hungry, slow access, mechanical HD of the last century, but... there is really nothing even close on the horizon right now :( Sadly, flash just isn't practical at all in it's current form for anythig OTHER than small devices that only need a small number of gig in a tiny form factor."

In a couple of Gig you can easily store an operating system, many applications and many documents. For company PC's it would make sense to just load the OS and applications from flash and store the documents on the network. Really big files -media files- may still be stored on a (external) disk, that spins up when needed.

Currently I am trying to create a (headless) server that just runs from flash, without any mechanical parts whatsoever (using a VIA EPIA mainboard, I don't need CPU cycles or high redundancy). With flash it will be silent, will use almost no power and quick to boot. Maybe I'll even try to use RAID-5 on a couple of flash drives, why not? RAID is rather fast when latency is low, so it should be possible to get rather high speeds even with flash ( 3 * 15 MB/s is still 45 MB/s - less than 60 MB/s for a hard drive, but close enough).

Re:hard drives are going away

[suv4x4]

I'm certain that hard drives will slowly go away to be replaced with Flash ram devices. As the price drops it will happen.

Let me correct the mistakes in your statement up there:

I [want] that hard drives [to] slowly go away to be replaced with Flash ram devices. Price drops [should] happen.

Just because Flash is better in your opinion than hard drives doesn't meant that prices will magically drop (a hundred times?) to replace hard drives.

Flying cars are also much better and have much lower latency but alas: it doesn't magically drive them into existence. Also your assumptions on the Flash reliability are also biased, if you used your (current state of the art) Flash stick as a full HDD replacement (booting up, swap files, temp files, and all that jazz) it'd likely fail much sooner as per the manifacturers' estimations themselves.

We're seeing a sound and argumented move towards hybrid drives. This is good. Don't blindly extrapolate trends: it's a mistake too many people do too often.

Would you say that RAM is going away in favor of gigs and gigs of Level 1/2/3 CPU cache? Boy I wish! But it ain't happening.

Re:hard drives are going away

[Anonymous Coward]

Let me correct the mistakes in your statement up there:

I [want] that hard drives [to] slowly go away to be replaced with Flash ram devices. Price drops [should] happen.

Good job! While correcting it you also invented a brand new language.

Re:hard drives are going away

[mrchaotica]

you can get 750 gig of HD for $350...

...in a 3.5" form factor. You can't get that for a laptop.

Sadly, flash just isn't practical at all in it's current form for anythig OTHER than small devices that only need a small number of gig in a tiny form factor.

In other words, such as on (thin-and-light or ultraportable) laptops. Although I agree it has no chance of replacing big discs on fileservers, desktops, or PVRs, I think it does have a decent chance of replacing them in portable machines.

Re:hard drives are going away

[Anonymous Coward]

f++?

Re:hard drives are going away

[AbRASiON]

I've been hearing speculation like this since I first built my first PC when I was 13.
That was 16 years ago.

Magnetic will always beat flash for physical amount of space and size, despite the unfortunate noise / heat / reliability issues.

Everyone knows

[Anonymous Coward]

Hybrids are sterile...

What I really want to know is...

[jhfry]

can the user control what is cached and what isn't?

For example, I could care less if a config file I will likely never edit again is cached, but I want my database to be cached for higher performance.

Re:What I really want to know is...

[drinkypoo]

The idea is that the OS handles this and automatically caches frequently-used files. But it's also used as a delayed write cache to keep you from having to spin up your hard drive due to infrequent writes (like log entries.)

Re:What I really want to know is...

[Dan Ost]

The flash in the hybrid drives won't be used as that kind of cache (you're thinking of the Vista's ReadyBoost).

This flash will be a write cache for the hard drive so that the hard drive doesn't need to spin up as often (this will potentially enhance your battery life). As you make changes to your data, it will be written to the cache and then flushed to the drive (a) when the cache is full or (b) when the drive is spun up for some other reason (a read, for example). Presumably, if the drive is already spun up, the flash won't be used at all and data will go straight to the disk.

Re:What I really want to know is...

[TheNetAvenger]

The flash in the hybrid drives won't be used as that kind of cache (you're thinking of the Vista's ReadyBoost).


Correct, this is not Vista ReadyBoost technology, it is Vista ReadyDrive technology.

Somehow people keep skipping the fact the write caching technology these drives are using is a MS designed technology, even though it is not ReadyBoost.

More info, try:
http://en.wikipedia.org/wiki/ReadyDrive

Or even www.microsoft.com

Re:What I really want to know is...

[Anonymous Coward]

I would like to see a nice big chunk of flash to hold things that aren't written too often, such as the kernel and executables. Then provide a hard-wired switch so that it's read-only unless the user intervenes by pushing the write button. That would not allow malware to overwrite anything by itself, but would still allow a person to update the system.

Re:What I really want to know is...

[0xABADC0DA]

Who invented the idea of an integrated disk read cache? Nobody knows because it's such a trivial idea that claiming credit for it just makes you look silly -- and desperate. If Microsoft is so stretched for innovation that they have to go around demanding props for "hey lets write these 100 bytes to flash instead of spinning up the drive" then they are in really bad shape. My advice is to jump ship now before the MS Titanic hits a penguin.

Re:What I really want to know is...

[TheNetAvenger]

If Microsoft is so stretched for innovation that they have to go around demanding props for "hey lets write these 100 bytes to flash instead of spinning up the drive" then they are in really bad shape. My advice is to jump ship now before the MS Titanic hits a penguin.


Go read about ReadyDrive and ReadyBoost and Superfetch before you make such an crazy assumption.

Do you really think HD manufacturers would be working with MS on such simplistic concepts if it were merely just a generic cache concept? I would bet HD companies have been doing more with Media caching than any other part of the industry, so for them to welcome MS research technologies into a product, it MIGHT just be more than a generic cache.

MS has some really good ideas with caching with ReadyDrive,ReadyBoost, and even the Superfetch in Vista, these are things the rest of the industry should be capable of realizing what IS GOOD about them and use that knowledge in other OS projects, instead of sticking their head in the sand and saying well it is from MS, it must suck.

Does everyone out here ignore their competition in all of their business activies? If you did you wouldn't have a business, but yet people in the OSS world have a BAD habit of doing this with MS.

Even if you hate MS with every fiber of your being, you can't ignore what they do and expect to remain on top in a field where they have more money than God to throw at R&D and Engineering minds.

Re:What I really want to know is...

[mrchaotica]

I think you're conflating two different (but related) technologies: the former function is designed to be used on separate flash disks that are about the same size as the system RAM; the latter uses 128-256 MB and is what these "hybrid hard drives" are for.

Re:What I really want to know is...

[drinkypoo]

Well, my understanding (which may have been a misunderstanding) was that you could get the benefits of superfetch with just a flash drive, but a hybrid drive gave you both...

Re:What I really want to know is...

[mrchaotica]

Heh, now you're conflating three -- superfetch is yet another separate thing. : )

Anyway, here's how it is: Vista has three technologies, superfetch, readyboost, and readydrive.

Superfetch preloads frequently-used stuff into RAM. It does not require any extra hardware, apparently. Readyboost basically acts as a cache between RAM and the paging file on the hard disk, so stuff that gets swapped out can get swapped in a bit quicker. It uses external (i.e., not part of a hybrid hard drive) flash, such as on a USB drive or in a CF or SD slot. Readydrive caches writes to the hard disk so that it doesn't have to spin up as often, which saves battery life. This is what's used with the hybrid hard drives.

These three technologies are related, but exactly the same.

cached database...?

[grahamsz]

I'd imagine you'd get a much better ROI if you invested in a suitable amount of RAM to keep your database indicies in RAM.

Hybrid

[alcmaeon]

When I saw this, I thought, "Cool, now I can finally get my kids a dog and me some additional storage space, all with one purchase."

Re:Hybrid

[Anonymous Coward]

When I saw this, I thought, "Cool, now I can finally get my kids a dog and me some additional storage space, all with one purchase."

And to celebrate your new found good fortune, you decided to take yet another bong hit. Peace out, man.

Solid state disks

[jones_supa]

I'd rather wait for the flash-only solid state disks to become affordable than buy one of these.

Re:Solid state disks

[Namlak]

I'd rather wait for the flash-only solid state disks to become affordable than buy one of these.

Yes, but buying one of these is what will help ramp up the production processes and generate the economy of scale that makes the flash-only drive you want more affordable in the future.

Old news

[Intron]

The press release from Samsung is dated April 2005. You can read more technical details there without all the annoying popups on ExtremeTech. Looks like the drivers which give the power savings were written by Microsoft. Planned ship date was late 2006, so they didn't fall too far behind.

Combine RAM with FLASH to store fs journal

[AaronW]

I would like to see a battery-backed RAM drive with FLASH as well. I think that for journaling filesystems it would be great for performance since the journal could be written into RAM and then later written to disk. The drawback of the RAM based drives I saw was that the battery is only good for a limited amount of time. The way to fix it is to provide less battery time but use that time to write the RAM out to FLASH when the power is cut. The advantage of combining RAM and FLASH is that RAM is very fast to write to and has an unlimited number of write cycles. Of course, I'd really like to see one of these new memory technologies come out that combines the best of DRAM and FLASH.

Re:Combine RAM with FLASH to store fs journal

[RattFink]

Another problem with battery backed ram is that without the overhead of a quite complex circuit to refresh it only works with the far more expensive, and generally only available in lower densities SRAM.

Re:Combine RAM with FLASH to store fs journal

[AaronW]

There are RAM drives available that use DRAM, but due to the refresh circuitry and whatnot it takes a bit of power so the battery will only supply power to the RAM for a limited amount of time.
Also, if the flash were removable (i.e. SD card, compact flash) then it could be possible to move to another machine.

Re:Combine RAM with FLASH to store fs journal

[developer55]

"The way to fix it is to provide less battery time but use that time to write the RAM out to FLASH when the power is cut."

Since the power has been cut, do you mean this battery to also power the hard drive that is being written to?

Re:Combine RAM with FLASH to store fs journal

[AaronW]

No. Since it is a journaling filesystem, as long as the journal is intact, all pending writes will be completed the next time the system is powered up. Journaling filesystems typically write data out to the journal first, before writing it again to the proper location on the hard drive and updating all the inodes. The idea is that if power is lost that fsck will play back the journal and complete all pending operations. The problem with journals is that two write operations are required. Some filesystems can speed things up by storing the journal on a different hard drive than the data. By using a ram disk, this should significantly speed things up. This does not work for all filesystems, though. Some, like Reiser 4, do not have the traditional journal and hence only need one write. EXT3 also has the ability to write both data and metadata to the journal first for the best recovery, but it makes writes much slower. Most just write metadata only to the journal for performance so after recovery the filesystem is in a consistent state but more actual data will be lost.

Re:Combine RAM with FLASH to store fs journal

[RattFink]

There are RAM drives available that use DRAM, but due to the refresh circuitry and whatnot it takes a bit of power so the battery will only supply power to the RAM for a limited amount of time.

I am well aware that RAM drives are available but that battery on that unit will last you between 4-5 hours and that is only for a maximum of 4 gigs. Once you scale that up to HDD sizes you would need another laptop size battery just to give you enough time to run from outlet to outlet. DRAM just wasn't designed for this kind of thing, is too leaky and inefficient.

Yeah, about Vista's Superfetch caching...

[empvirus]

I used to think that it was a good idea, up until I saw it go horribly wrong on my friend's laptop. The thing bluescreened, then wiped out all partitions on his HDD, including his EXT3 linux partition, and the contents of his flashdrive were completely gone. He couldn't recover ANY of the data he lost. And last I checked, he's still having troubles with the MBR and grub. I know, this is anecdotal, but I thought at least one or two people would like to know that Microsuck still has a ways to go before their little feature is usable.

I smell BS

[cybrthng]

Superfetch will shut itself down if it fails. It couldn't possibly cause a FS crash/corruption like you complain.

OTOH these drives could fail since they're not superfetch and they're potentially caching writes.

Re:I smell BS

[drinkypoo]

Superfetch will shut itself down if it fails. It couldn't possibly cause a FS crash/corruption like you complain.

Famous last words: "What could possibly go wrong?"

Wait for Intel PRAM

[Nom du Keyboard]

Rather than ship hybrid drives now with flash chips good for a few thousand cycles, why not wait until the end of this year and ship them with Intel PRAM or equivalent. PRAM is expected to be faster, non-volatile, and handle many times more R/W cycles. Or is the lifetime of the rest of the drive no longer than for the flash itself? This seems to be to be just a bit ahead of its time, and has the potential for either problems, or performance degradation, over a relatively short timespan.

Re:Wait for Intel PRAM

[Anonymous Coward]

A current-generation flash drive can support ~7 years of constant writing.

Re:Wait for Intel PRAM

[njchick]

Some company waited for 802.11a chips instead of releasing 802.11b cards. That company is not in wireless business anymore. Sometimes it pays off to be first with an inferior product and then offer incremental compatible updates, rather than wait for the perfect solution and have your product perceived as offering little difference at the price of incompatibility with competing products already on the market.

Different than the drives designed for Vista? Not.

[TheNetAvenger]

I'm not sure what is more screwed up the article linked to about the drives or the Slashdot comment.

ReadyDrive is NOT ReadyBoost, but it IS STILL a MS Technology and is designed to work directly with Vista.

http://www.microsoft.com/windows/products/windowsv ista/features/details/performance.mspx

Also why does the linked article and Slashdot dismiss these drives as having nothing to do with Vista, when in fact they were DESIGNED Specifically to be used with Vista and employ MS Vista technology in the hardware?

Is Slashdot trying to become the misinformation site of the Internet?

http://www.digitimes.com/systems/a20070307PR201.ht ml

http://www.channelinsider.com/article/Samsung+Ship s+Worlds+First+Hybrid+HDD151or+Is+It/202621_1.aspx

"Optimized to work in Windows Vista-capable notebook PCs, Samsung's MH80 is a 2.5-inch hybrid hard drive with 128 or 256MB of flash memory. It combines a hard disk drive with a OneNAND Flash cache and Microsoft's ReadyDrive software, offering faster boot and resume times, increased battery life and greater reliability compared to traditional magnetic media technology, the spokesperson claimed. "

Sorry slashdot, but these drives are designed for Vista. Sure they may offer performance improvements in other OSes, but will see the majority of performance gains in Vista. Also even when used with other OSes, the way the Drives internally manage the Flash caching is from MS, so thank them the next time you boot your Linux laptop with one of these drives.

As for the other questions people have about the limited write times of Flash RAM, etc, go lookup MS Superfetch technology which specifically addresses these issues by writing to various locations in the Flash space, since this this is also how these drives work to ensure the same bits don't always get used, giving the flash cache the equivalent or greater lifetime than the HD platters.

I know this is SlashDot, but someone could get the fact right once, right?

Are you gonna go my way?!

[VisiX]

Sorry, had to do it.

Hybrid Drives Rock

[Grashnak]

They really cut back on emissions by only using the gasoline engine while they're spinning up, and then switching over to cleaner battery power while slowing back down.

Even that is not quite true...

[Anonymous Coward]

It's more like 1,000,000 ERASES. You erase an entire block at once (set it all to 1's) and then you can write to whatever parts of it you want by changing some of the 1's to 0's. You just can't change them back to 1's without erasing the entire block (which is usually something large like 64K).

For those who were paying attention, this means that if you are careful, you can log lots of small things into the same block, and never erase it until it gets full. I know a guy who got a patent years ago for RIM on a log-structured filesystem in flash, that was designed to take advantage of this property to the max. I think you could write continuously to the blackberry's filesystem for like a decade or something, without wearing out your flash.

Re:Different than the drives designed for Vista? N

[jschoenberg]

Totally agree with you! I can't believe that the editors didn't catch this amazing amount of misinformation before posting it. This points out the root problem with internet media....lack of editorial input means that fact checking just doesn't happen. Media without facts is a sign of the apocalypse!

Just a side note

[wellingj]

I personaly wouldn't get one of these. I use my laptop for coding and DVD watching.
For me it makes better sense a full solid state flash drive because it uses less
battery life and is probably a little quicker and more quiet. Of course if you need more than 8GB
of storage, the price is a little prohibitive. That's why I use SVN, and store all my code at home.

Abstractions inside the controller

[tepples]

Perhaps the difference between NAND flash parts rated in erase cycles and NAND flash parts rated in MTBF or MTTF has something to do with abstractions inside the controller. Flash "chips" commonly use a bare-bones interface like that of SmartMedia, while flash "drives" have an ATA, USB, or SD controller in front of the flash that performs error correction and wear leveling. I'm pretty sure that's where the 5% difference between a 512 mebibyte underlying capacity and a 512 megabyte actual capacity comes from.

ECC

[tepples]

What that means is that statistically speaking, there's a good chance that most of your chip will still be writeable up to a million times. Some bits will fail sooner, some will fail later. But that's what turbo codes are for, right? You give up a percentage of capacity in favor of reducing the data's bit error rate to epsilon.

Imagine 20 miniSD cards in a RAID-5

[tepples]

you can get 750 gig of HD for $350, probably less now, how much would that cost in flash? For desktop-replacement applications that need more than half a terabyte, such as video editing, hard drives are probably the best option. But with fully-packaged flash retailing near $10 per GB, a laptop with a flash drive (imagine an enclosure the size of a 2.5" hard drive containing 20 miniSD cards in a RAID 5) can do a lot of things surprisingly well.

Sadly, flash just isn't practical at all in it's current form for anythig OTHER than small devices that only need a small number of gig in a tiny form factor. Define "small number of gig" in terms of applications that laptop owners would want to run and which wouldn't work with a "small number of gig".

Flash RAID

[tepples]

With a hard drive you can get high transfer rate and high storage capacity. With flash you get low latency and low power consumption. With flash you also get small physical size. If you don't care much about size, you can get high transfer rate by RAIDing a whole bunch of miniSD cards.

PSRAM

[tepples]

Another problem with battery backed ram is that without the overhead of a quite complex circuit to refresh it only works with the far more expensive, and generally only available in lower densities SRAM. There is PSRAM, a form of DRAM that has the refresh circuitry on the die. It can achieve densities comparable to DRAM because it is DRAM.

Re:PSRAM

[RattFink]

The overhead is still there just because it's on the die doesn't make the overhead go away, because of that battery life on PSRAM is much lower then that on battery backed SRAM. As for density, it's in the same boat as SRAM, while higher densities can be made, they typically aren't and you are stuck with in most cases a maximum density of 64-128mbit, much lower then DRAM.

Why isn't flash faster?

[mrcaseyj]

Does anyone know why USB and IDE flash drives don't max out their bus bandwidths? I realize that a flash chip can only go so fast, but why don't they just parallel as many as needed to get the desired bandwidth?

I'm in the southern hemisphere you insensitive ...

[gfim]

Why say "last fall"? I realise that Slashdot is mostly full of yanks, but why not try to be a bit more cosmopolitan?

Re:Different than the drives designed for Vista? N

[Trogre]

So what? Most PC hardware is designed specifically for some Microsoft software (DOS, Windows 95, Windows XP) and has been for the last 20 years. Doesn't stop you accessing that technology with a Linux kernel.

Re:Different than the drives designed for Vista? N

[TheNetAvenger]

and has been for the last 20 years. Doesn't stop you accessing that technology with a Linux kernel.


Very true, but for Linux to take full advantage of caching in a hybrid drive, it needs to also alter the memory management, caching, and paging techniques Linux uses.

The drive is going to transparently provide a boost in performance for any OS, but when used with Vista, the direct management of when and for what to use the hybrid cache for is something the OS is already designed to do. For example Vista knows what to put where based on flash based performance differences that are both slower and faster than the HD platters. So random location data gets a priority for Flash, but sequential data skips the Flash portion of the drive as it cannot give the same performance as a HD in reads, and sometimes writes.

The drives inherently (from my understanding) do this to a degree, but Vista adds in the intelligence of what data is cached based on past performance and the expected usage of the user. Vista can optimize the basic functions the internal HD is doing on a more basic level based on specific needs of the FS, OS, and User.

Anyone that is interested in how Vista uses Flash caching differently than other OSes, should check out ReadyBoost and ReadyDrive a bit more in depth, MS provides fairly detailed technical articles on what is happening and why it works the way it does.

You can also find information at some of the HD sites that are coming out with Hybrid and full Flash based Drives. For example, the full Flash devices are obtaining sequential read/write speeds equal to platter based HDs by using an internal Flash RAID technology.

A lot of this is fairly interesting as we are finally getting to the place were moving storage devices will slowly start to disappear.

This is also another reason I tell my friends in the OSS world to pay attention to what Vista is doing and promoting in terms of basic architectual technology. There are plenty of things to beat up Vista about, but there are a few things Vista is ahead of the industry and deserves the attention of movers in the non MS world so that Vista doesn't move ahead with unforeseen advantages.

Hybrid drives and Flash caching is one area MS has done their homework; GPU scheduling and GPU virtualization is another area that could isolate technologies and hardware to Vista if others in the industry don't start paying attention now.