Samsung's Hybrid Hard Drive Exposed

Erica Campbell writes "Samsung is preparing to release a new Flash memory-assisted computer hard drive that boasts improved performance, reduced energy consumption, a faster boot time, and better reliability. The new hybrid hard drive will be released around the same time as the upcoming Windows Vista operating system and will be one of the first hardware designed specifically to benefit from it."

What's so special about Vista?

[CastrTroy]

What's so different about Vista that makes this drive benefit from Vista. Will the drive not work in Windows XP, Linux or Mac OSX machines?

How long until failure?

[Anonymous Coward]

Sadly, I can't RTFA as the account has already suffered a slashdotting, but I have a question about this drive. Isn't there an upper limit to how many times you can write to flash memory before it ceases to function? Granted, hard drives wear out eventually, but unless this stuff is of high quality then the cache is going to wear out before the rest of the drive.

When the cache dies off, what happens?

Linux Next?

[Yehooti]

It's designed for Vista, but I want it for Linux. How long until then I wonder?

Apple?

[SuperBanana]

The new hybrid hard drive will be released around the same time as the upcoming Windows Vista operating system and will be one of the first hardware designed specifically to benefit from it."

Given Apple's strong relationship with Samsung (iPod shuffle+nano memory both come from Samsung, I believe- and I'm almost positive Samsung has supplied RAM to apple on+off since the golden olden days), what do others think about the possibility of this ending up in a Powerbook, er, Macbook Pro- and 10.5 being designed to take advantage of it?

Apple can be hit or miss with the latest and greatest- they took forever with USB2 (yeah yeah, firewire blah blah) and lagged behind a lot of the smaller laptop mafacturers with Expresscard (given there's next to nothing for expresscard, who can blame them?)...it'll be interesting to see if Apple thinks this is a win or lose technology...

Re:Ship time

[Who235]

Excellent point. TFA has been Slashdotted so I can't see the specs without an undue amount of initiative on my part, but presumably if they put enough flash memory in, and used a proper distributive algorithm to make sure the same sectors aren't constantly written to - there might be little danger of this drive failing any sooner than a conventional drive.

I hope.

Re:How long until failure?

[bcat24]

I think the cache is designed to help with booting and suspend/restore, so it shouldn't be written too much. With a large enough flash buffer, it should be able to least for the normal life of the drive.

Re:Linux Next?

[JimXugle]

You can have a similar effect now by using a flash drive as your root partition, or as a swap partition. Keep in mind that using it as a swap partition would make the drive age faster.

Re:What's so special about Vista?

[MindStalker]

Vista is designed to be bootable from flash memory. Significant changes to the bootcode of XP would be nessesary for the instant on features. The other features could possibly be incorperated with drivers.

Why does the CF have to go on the disk?

[pensivepuppy]

If Vista knows about the CF, why does it need to be on the hard disk itself? It sounds like all the heavy lifting is being done by Vista anyways. WOuldn't it make more sense just to use any CF attached to the system for this caching, etc, and use normal hard disks instead? That way adding CF to a PC would improve its performance, no matter what type of hard disks you have attached.

Re:What's so special about Vista?

[timeOday]

As a Linux laptop user (yes, there are a few of us) super-fast bootup would be a very attractive feature, and an advantage now falling to XP. I'm curious how the boot time will compare to a resume from "suspend to disk" (though the attractiveness of suspend to disk / suspend to ram are limited by the fact that they're often a nightmare to set up anyways).

Re:Why does the CF have to go on the disk?

[Achromatic1978]

I was under the impression that it did. I plugged a CF card into my laptop when it was running a Windows Vista beta, and it popped a dialog along the lines of "Do you want to use this device for file storage, or for increased system resources?"

Re:What's so special about Vista?

[NineNine]

That is actually reason enough for me to re-think that whole Vista thing. With partial flash drives and eventually 100% flash drives, the last major component of computer hardware failure, namely, all of those closely moving parts in a hard drive, will be wiped out. Wow. That sounds pretty cool.

Oh yeah, and it'll be fast as hell, too.

Re:What's so special about Vista?

[chmod a+x mojo]

What laptop are you using? My compaq had no problems with suspend to ram/disk (i did have to set the options in the config files to suspend anyways), possibly it is because it has a AMD processor in it. I normally prefer intel but it was cheap and gets the job done quite well. Maybe AMD has better ACPI standards compliance? My intel p4 northwood is definatly NOT acpi compliant ( it does not even have the full ACPI instruction sets according to a few inux distros )

Re:SuperFetch uncool...

[dbIII]

I'm sure that Vista is smart enough to free up the RAM that SuperFetch is using if it could be better used for something else.

It's a tweaked XP not something newer than plan9 - it will probably swap it out to disk so you get a big page file and a delay while it is doing it, which is probably one reason this new drive will help.

Personally I think it is stupider than doublespace since memory limited programs like image editors are commonplace now. The annoyance of not being able to print for a couple of minute while the memory swaps out can come to everyone. It is paticularly stupid on the MS Windows platform where people typically only perform one application task at a time - you do not want the behemoth that is office in memory while you run some graphicly intensive thing or vice versa - you almost always only have one user with one desktop and one application being worked on.

Re:Flash

[Tatsh]

The problem here is if we're talking about fully flash hard drives. If you bought two, one for backup (even in a ghost or RAID ghost setting), how can you rely on the fact that one won't die when it was written onto the same exact number of times, byte for byte, and used for the same amount of time? It's similar to current HD's, but these tend to not break at the same time.

Flash also can only be written onto so many times before it's rewriting capabilities start to suffer badly (don't remember the exact number, but this is when flash drives die).

Has it changed in 3 years?

[Bill, Shooter of Bul]

We wrote a driver to read and write fat 16 flash drives for an embeded system. The testing for it wrote and read full speed 24/7 for two weeks before they died. I assumed that was because of the limited read write settings. Or is it possible the low quality connection was to blame? Doesn't really matter they were only used to transfer settings. As any one whos had to support them knows, they often just die for no apparent reason. I'm not convinced that this is a system I'd want my data on.

funny?

[oohshiny]

I can't tell whether you're trying to be funny.

That is actually reason enough for me to re-think that whole Vista thing.

Has the ability to boot and run Linux off flash made you "re-think that whole Linux thing"?

With partial flash drives and eventually 100% flash drives, the last major component of computer hardware failure, namely, all of those closely moving parts in a hard drive, will be wiped out.

They'll be replaced by a medium that has a much higher MTBF for writes.

Oh yeah, and it'll be fast as hell, too.

Not really. Flash memory is not all that fast, and a lot of boot time is spent doing other things. On all my machines, most of the booting process is concerned with checking and initializing hardware.

Re:What's so special about Vista?

[anethema]

No, but flash DOES fail fairly predictibly and in certain ways. You get a certain amount of write cycles per sector, etc. You implement a write spreading alg on the drive and increase its life dramatically. You can easily start marking sectors as bad and have VERY early warning on drive failure. You can extend life dramatically by having extra sectors on the drive for write spreading (more benifet the more full the drive is). Also with current write lives, we can have drives you can write to 24/7 for years without that sector failing. With the write-spreading the drive lifetime would vastly outstrip a normal hdd on average.

The access time is also VERY low compared to a HDD, and unless the controller itself fries, its almost impossible to have catastrophic data loss.

Basically, we cant switch fast enough, there are no downsides but price.

Fixing software problems in Hardware

[Casandro]

Actually I think it's more like fixing software problems in Hardware. The situation in which this technology will improve access time is when you have to randomly seek on your harddrive. Unfortunately that is needed in Windows as there is little possibility to keep all your bootup files one after another in the order you need them. With Linux, however that is rather easily possible. You can create an initial ramdisk which the computer can load very quickly without much booting and then boot from it. Theoretically, if you have enough RAM, you can even load your complete system into it.

So now you essentially have to spend a lot of money (Flash and Patents!) on a technology which will, at most, give you an decrease in boot-times and will be obsolete once the power management of the drivers support Suspend to Disk or Suspend to RAM. Just look at Linux or MacOS 8 on an old clamshell iBook. You just close it and it's "Off", you open it again, and after very few seconds it's completely back again.

Re:Ship time

[schmiddy]

Out of curiosity, how similar is Vista's ReadyBoost feature to just mounting a USB drive as swap in Linux? Can you hot-unplug the drive in Linux if it's being used for swap? According to a FAQ on ReadyBoost [msdn.com] I found, Vista will back up the pagefile to disk so it's not a catastrophe if you yank out the USB stick.

And is there any setting in Linux to tweak to let the system know you've got a fast swap partition, other than simply monkeying with /proc/sys/vm/swappiness ?

Flash has gotten better, but it can still fail...

[figgypower]

Flash has improved by leaps and bounds, with the earlier Flash drives supposedly lasting 1000 read/writes. Nowadays, from what I can gather the average Flash drive lasts 300,000 to 1,000,000 erase/writes. Good, but still finite and, relatively speaking, more finite then HDs. Flash also has cycling quality control problems, which are not as severe in the HD industry. All that said, I'm not saying Samsung shouldn't go ahead and make its hybird hard drive, but consumers should realize that the Flash component won't necessarily have the necessary longevity. This will proably be true with Samsung drives, because they're not likely to place small init files onto Flash but cache larger data (after all, what's the point of caching these small files if the performance improvement is going to be so low and just one time).

Sources:

http://www.techworld.com/storage/features/index. cfm?featureid=2814&pagtype=samecat

http://www.bitmicro.com/press_resources_flash_ss d.php

http://en.wikipedia.org/wiki/Flash_memory#Limita tions (yeah, yeah, Wikipedia isn't authoritative, but it's good enough)

Re:Linux Next?

[keesh]

Except that Linux does preemptive swapping long before you run out of RAM, so that if something suddenly needs it, there's no huge delay whilst things get swapped out.

Use Flash for directory structure not cache!

[Anonymous Coward]

We did some basic research with Flash / HDD hybrids two years ago. As such disks weren't available, yet, we were using a real (Notebook)HDD and a IDE-Flash-HDD in parallel.
Our goal was to minimize energy consumption for mobile devices (i.e. not a lot of ram available for caching and the device is switched off repeatedly to save energy).

Using a very sensitive (time resolution wise) energy measurement device, we determined, that most energy was consumed by moving the heads into position. The difference was substancial: Around 0.63W for the HDD spinning idle and about 5.3W during heavy seeking (e.g. trigered by a "find ." in the root of a freshly booted system).

We decided to not use the flash as cache (flash is quick to read, but slow to write) and just put the relatively static metadata (directory structure, inode tables...) onto the Flash drive, but keep there files and data on the HDD, as each directory access triggered a expensive seek, but delivered very few data, compared to reading a file.

To simulate our mobile device we used a Linux-System limited to 32 ram to prevent the system from excessive caching.
We observed up to a factor 8 reduced energy consumption and as a surprising side effect a factor 6 increase in speed!

When increasing the available Ram, this advantage quickly vanished on repated benchmark runs, as the System appearently cached the directory structure very effectively. The first run after booting however still performed substancially better with our system, no matter the amout of ram. (And this was our target useage profile: Power on, search something, Power off).

As the code was an embarrassingly ugly hack to the ext2 driver and we envisioned trouble keeping the hdd with the data and the flash-hdd in sync, it was not persued further.
However with hybrid drives becoming available, it might be worth a more detailed analysis...

Re:Flash parts die in 3 weeks of solid write tests

[Tim Browse]

True, a megabyte read can be fast in flash, but lots of random 512 byte reads or writes are far slower than a modern hard drive STILL in 2006. (15,000 rpm scsi from 7 diff manufacturers for example).

Really? What are the seek times on these 15,000 rpm scsi drives that out-perform solid-state devices, and where can I buy them?

Re:Linux Next?

[Tesen]

You still have a physical disk inside that is spinning, the drive heads still have to wait to reach position to begin read. If your drive is fragmented badly, then there is additional seek times also.

/boot?

[TheCoop1984]

If the flash part of the drive can be accessed seperately to the platter part it would seem like an ideal place to put /boot or /lib/ld-linux.so etc - libraries that are loaded on every boot. This could seriously speed up the boot time, the disk might not even have to spin up at all during bootup.

It would also be ideal for laptop systems to save power...

Re:Linux Next?

[misleb]

Except that Linux does preemptive swapping long before you run out of RAM, so that if something suddenly needs it, there's no huge delay whilst things get swapped out.

Monitor swap usage on any healthy Linux system and you'll notice that usage remains pretty low most of the time. Anything you DO swap out should be largely unused anyway. So I don't really get how using flash for swap woudl wear it out faster unless you were using it as a substitute for more RAM.

-matthew

Re:Ship time

[dgatwood]

Enable swap on a flash volume. Install it in your grandmother's computer (256MB RAM, with the OS recommending 512MB). This is a real cause for concern, IMHO.

Not all wear leveling algorithms are created equal. In the worst case, your writes get hashed across a relatively small percentage of the cells in a particular flash part. Even in the best case, your wear leveling is still per flash part, not across the entire hard drive. Thus, unless the drive is spreading the cache load out evenly across all parts (in a way that is persistent across power cycles for the drive), it is really easy to construct a case where you would artificially wear one part of the unit faster and start getting errors when the drive is much less than 3-4 years.

The issue has been partially worked around, but the workaround is only sufficient for devices that don't have heavy write loads (e.g. digital cameras), not for a main system volume or even necessarily for a cache, though it -might- be good enough for a cache. Without a lot of implementation details, I wouldn't be comfortable with this feature on a hard drive. Thus, I'm not going to jump on this bandwagon until they've been out there for several years just so I can see the failure rate figures before making that call. I also want a guarantee that I'll be able to pull or install a jumper and disable the flash entirely so that when it eventually fails, the drive will still be usable, albeit slower.