Slashdot Mirror
Seagate Launches Hybrid SSD Hard Drive
MojoKid writes "Though there has been some noise in recent years about hybrid storage, it really hasn't made a significant impact on in the market. Seagate is taking another stab at the technology and launched the Momentus XT 2.5-inch hard drive that mates 4GB of flash storage with traditional spinning media in an attempt to bridge the gap between hard drives and SSDs. Seagate claims the Momentus XT can offer the same kind of enhanced user experience as an SSD, but with the capacity and cost of a traditional hard drive. That's a pretty tall order, but the numbers look promising, at least compared to current traditional notebook hard drives."
The cache is 4GB, the drive is up to a 500 GB 'traditional' drive.
I see. Then I must leave.
Disagree != mod troll.
What swapfile? I have used Ubuntu on a few PCs with at least half a GB of RAM, and I rarely see swap usage climb above 40 MB. In an environment where reads are cheaper than writes, you'll want to use a low value for the swappiness, such as 10% instead of the default 60%.
Hybrid storage drives should be manually manageable.
You should have the possibility of configuring which files/folders/partitions/whatever you want to be accessed fast and which parts are to be left as "long term", slow access, storage.
I wonder if this is simply a more expensive version of ReadyBoost. Similarly, it takes your most frequently used files and puts them on a flash drive for faster access times, in a way that is transparent to the end user. In this case I wonder if there would be any speed gain from using this on a PC running Windows 7 with ReadyBoost? Caching always introduces some overhead, so rather than using multiple levels of "flash cache" it might be better to simply turn ReadyBoost off in that case. My experience with ReadyBoost has been that it does indeed improve performance, but in no way close to using a real SSD as the system drive.
Quality, performance, value; you get only two, and you don't always get to pick.
OCZ and im sure others have SSDs up to 500GB now. OK, they cost as much as my car, but they exist. It wont be long before they get up to 1TB, then 2TB.. Then its just a matter of waiting for the price to come down.
SSDs have caught up to traditional drives capacity extremely quickly, it wont be long before you can put a 10TB SSD in your laptop and never have to worry again (well, except for loosing it).
These days with RAM being so cheap, your swap space is basically a warning that things are going terribly wrong. You want your swap on slow storage because slow storage is cheap and your swap should see very few writes under normal operation. If your machine starts hitting swap like crazy, you'll know immediately because your performance will go straight down the crapper as it feverishly tries to write to slow storage. This is your cue to figure out what's wrong and fix it ASAP so your machine will stop thrashing.
It wouldn't help start up time would it ?
New things are always on the horizon
The trend continued in the remainder of the PCMark Vantage tests. On the first run of the benchmark, the Seagate Momentus XT performed surprisingly well. By the third run though, performance had increased dramatically and approached the level of the true SSD.
We should also point out that we ran this test numerous times, and after the third run, the additional performance increases stopped, which is to say Run 4 performed like Run 3. The screenshots of the actual Vantage performance summaries are available in the image gallery at the bottom of the page for those that would like to see the progression from Run 1 through Run 4.
So, it is slightly more expensive than a high performance disk drive, and offers most of the performance of an SSD. Most room on hard drives is taken up by massive media files, which do not need to be accessed at top speed because they are usually streamed for playback.
Eventually the the best drives will allow you to designate a folder for SSD storage only. Video editors should be able to buy a 1TB/32GB SSD drive and have a folder for the files they are currently editing. This may not be necessary if the drive intelligently identifies open files and transfers them to the SSD portion.
And I don't think this is just a big cache. I'm pretty sure hard drive caches disappear upon reboot.
According to the article, it's SLC flash. It should have many more write-erase cycles than MLC.
The difference between this drive and every other drive on the markt is not the fact it has a cache. Every harddisk has a cache. The difference is: this drive has a 4 GB cache. And because it's an SSD cache and not (as you suggest) a RAM cache, it maintains it's state even between reboots, so your computer is fast right from the start.
Pretty good is actually pretty bad.
No, because then every time your computer reboots, you need to fill the cache again. Using a solid state cache, you need to fill the cache only once and then keep it updated according to your usage, but a reboot wont harm performance at all.
Pretty good is actually pretty bad.
The main thing I use swap for these days is hibernating my laptop. What I need is persistent storage - the quicker the better.
Only if you want the SSD to die within a month. Windows writes to the swapfile pretty much constantly.
Oh dear, I'd better remember to replace my SSD-based laptop, err, 9 months ago.
Seems like you could do better if you simply could reorder the files on your traditional hard disk so that you'd get 100% readahead buffer hits. If properly optimized this way, your traditional hard disk should always be transferring near the max block read rate of ~100MB/s
I'm guessing this is what some of the boot profilers / optimizers are doing.
The readahead utility used by Redhat / Fedora (and also available for Debian) gives you some benefit loading lots of small files from the disk by reordering reads by inode number to minimize head seeks. The next major benefit would be if it could actually reorganize all those files into a single tarfile, and maybe even compress it a bit, so it can do a single large block read to get all that content off of disk and into RAM cache.
Filesystems have a much better idea of what data is going to be used frequently. This is an optimization they should be making. Seagate can make some good guesses by looking at block-level IO statistics, but that's like trying to optimize bytecode, all the really useful information is gone by the time you get to block-level IO.
I think hardware vendors should be supporting more interesting experimentation on the filesystem front instead of coming up with proprietary hacks like this that are basically a half solution.
Need a Python, C++, Unix, Linux develop
Ooh, good idea. By the way, I've been thinking of making an alarm clock that electrocutes your testicles if you hit the snooze button - can I sign you up for the beta?
If you were blocking sigs, you wouldn't have to read this.
They haven't been omitted. You need to read the second link. IOPs and read/write speeds are about 1/3 and 2/3 of the way down the page, respectively.
There is no "-1 offended" or "-1 you don't agree with me" mod options for a reason.
The way to get both performance and storage right now is to by TWO disks. An amazing concept I know. Who would have thought it was possible to get more then one HD/SSD into a PC.
Every single story about SSD's seems to bring out the idiots who want everything on one disk. Good thing these guys ain't farmers or they would be trying to plow the field with a Ferrari or cruise town with a tractor.
This drive is only of use to people who can't afford a real SSD and are limited to a laptop with only one drive bay and even then you would get far better performance with a normal SSD and an external drive for your porn collection.
Yes yes, there are people who use a laptop AND have need for far bigger datasets but on the whole, those people also need far greater access speeds then a traditional laptop HD can offer. I find it amazing to see someone claim he needs to edit video on a laptop with a 500gb 2.5 inch HD running at 5400 rpm. Who are you trying to kid?
And this drive won't be much help here. 4GB is just a cache file, if you are lucky it caches the right files but if you are doing complex stuff these "smart" caches often get horribly confused and start caching the wrong data. Like Vista trying to cache torrented files. Yes, I know it accesses the file a lot but please don't try and cache a 10gb file on the same HD. What's the fucking point? If you for instance will be running a large database from this drive, I am willing to bet its cache performance will degrade as it simply has to much to cache. Small caches only work when a small amount of files is requested a lot and the rest isn't. Like a porn collecton on your OS drive. Video editing, databases, filesharing always screw up caches.
If you really want performance in a laptop, spring for one with two drive bays, put as much memory in it as it can hold and get an SSD and a HD. A real SSD not one of the cheap ones some laptop companies put inside. An SSD is NOT just a fast HD, they truly are in a class of their own. And even if you got only a small single SSD, then you can still save space by putting your music/porn on a flash card or usb stick instead.
I wonder if people can ever get it into their heads that an SSD is about speed, not about capacity. Then again, since every single netbook these days comes with a 360gb slow ass HD instead of small but fast SSD, I think I might be fighting a loosing battle. Seems the average customer can only judge something if the number is bigger.
MMO Quests are like orgasms:
You may solo them, I prefer them in a group.
Yes
This drive still suffers from the historical bugaboo of spinning platters: it is damaged by shock. Also, it has the power draw (and heat output) of other spinning media.
Those are the two biggest reasons for SSD, especially in notebooks. Performance improvements are a factor, but I think they're the least interesting. In this respect, Seagate still needs to bring an answer, and they need to do it fast to justify their run up in stock price.
--
$tar -xvf
Solid state drives are flash-based and may never have the price:GB ratio that mechanical drives have. Methinks that even when 1TB SSDs become relatively cheap, mechanical HDDs will still reign in price and long-term reliability for heavy data usage. The deal with this hybrid product is that it combines both technologies to get the best of both worlds today. 4GB of flash is cheap, and obviously, so are mechanical drives. Combine them to get similar performance to a SSD and reliability of an HDD.
What makes this special is not just that it has a cache. Every HDD out there has a cache. This puppy has a "cache" 100x what current drives have. What's more is that this cache is persistent/nonvolatile. It's good when you reboot, so even at OS load, you see the advantages.
I have one of the old Samsung drives from 2007, a 80GB with 256MB of flash.
http://en.wikipedia.org/wiki/Hybrid_drive
http://www.samsung.com/us/consumer/office/hard-disk-drives/hybrid-hdd-flashon/HM08HHI/index.idx?pagetype=prd_detail
No, I would expect a harddrive to work in Linux. A harddrive which relies on ReadyDrive would not be a very good product, as it would only work correctly in Windows. That's why those type of harddisks never caught on, even though Microsoft did try to push this concept.
What Seagate is doing now, is using the ReadyDrive-concept of hybrid harddrives, but provide ReadyBoost-type technology on the controller of the harddisk instead of relying on the operating system.
Pretty good is actually pretty bad.
Yeah, that wikipedia article is using figures from over a decade ago (1997 for their highest NAND numbers). Today the number of write cycles for the premium chips are in the millions, not thousands. You'd have to write your entire 500gb hard drive 8,000 times to burn up that 4gb of flash, assuming 1 million write cycles. In these applications, though, they're probably using the "creme de la creme" of flash chips, which can last up to 5 million write cycles. That lets you write that 500gb hard drive 45,000 times before the flash dies. That's 22.5 petabytes worth of data transfer. Assuming an 11ms write time for the hard drive (typical of notebooks), and assuming I didn't royally screw up somewhere (I may have), I get 32 million years of non-stop writing to use that up.
Even at their slowest, SSD drives are 5-10x faster than spinning disks. This allows it to act as a perfect layer of cache between RAM and HDD. It's also non-volatile, like the rest of the hard drive, so all you need to ensure no loss of data is to at least get the data to the cache. Assuming the hard drive doesn't do something retarded, like wipe the cache on reboot, it's all good.
Security is mostly a superstition... Avoiding danger is no safer in the long run than outright exposure. - Helen Keller
Swap tends to be demand paging, where you're loading in a page as it's accessed (and writing a page out to make room, although this happens asynchronously with a little buffer). Satisfying page faults from a drive with a 5ms access time limits you to around an 800KB sustained transfer rate for reading. If you're lucky, you can do sequential writes so you keep your 800KB/s write rate. At this rate, you'd only use 7GB/day on swapping, if all of your disk I/O is swapping and you are swapping constantly for 24 hours. If you're writing more than a few hundred MBs per day to swap then you are going to notice that your machine has really terrible performance.
I am TheRaven on Soylent News
Actually, since semiconductor manufacturing gets cheaper practically exponentially (yay Moore's law) eventually SSD prices should catch up and undercut mechanical disk prices, just because of the manufacturing process. But that's a long ways off yet.
Well... maybe. NAND flash isn't too many generations away from hitting a lithography wall from what I remember. And we're still 10x-20x more expensive then magnetic media on a $/GB measurement. 2.5" laptop drives are in the 0.14-0.17 $/GB range, while SSDs are still up in the 1.36-2.34 $/GB range.
Assuming that we're already down in the 40-50nm(?) range for flash media, can they really manage to squeeze one more order of magnitude out of the feature sizes? (Closer to 3x smaller features since you double the bits at sqrt() feature size.) If they can make NAND flash down at 10-15nm, then yes we might see $/GB prices below $0.20.
25nm is apparently around the corner, so we might see a 2x-4x price improvement soon. The question is how much headroom is left after that.
Wolde you bothe eate your cake, and have your cake?
You're forgetting one thing:
Sometimes, a machine will go from seemingly normal to suddenly thrashing about in swap rather heavily, with no warning at all. This has been the bulk of my experience, anyway. When your machine gets to that point, and you're in a graphical environment like the majority of desktop users are, you may not be *able* to look into the problem at all. You have to wait until after the damn thing comes to its senses, because you can't even switch to a regular text console, let alone log in in from another box. Forget trying to spawn a terminal. Every little program you launch to try to find the cause just causes the machine to use more memory or swap at this point, which just compounds the problem.
When the offending program finally does end, it's too late to see what went wrong because most programs leave no traces of their actions other than doing whatever they're programmed to do. Unless you're running some kind of process/resource logging program on your box (I'm not aware of anyone who does this outside of security professionals perhaps), good luck finding out what actually caused the problem, unless you saw something visibly bug out just before the machine stopped responding.
There are no two ways about it - this is absolutely the worst way to handle an out-of-memory condition. Most people would much rather have programs complain about lack of memory than to have their machine "lock up" for an hour while sitting there churning away in swap. In my experience, the average user figures their computer's being stupid again, and it's time to hit the power switch or the reset button, or maybe call someone for help (which doesn't work anyway, so they're back to square one).
To give an example, I once set my machine off to run a build which should have taken maybe half an hour, and went off to run some errands and watch a movie. It was still going three hours later, and had dug the machine so deep into swap that mouse events were taking 10-20 seconds just to echo to the screen, and keyboard events were nonexistent as far as X was concerned. I tried my level best to bring the machine back to a sane state, but I eventually had to give up and hit it with Alt-SysRq-U/S/B.
I love Linux as much as anyone, but I got sick and tired of this happening on my boxes, and responded the only way that seemed to make sense: I disabled swap entirely on both systems and added enough RAM to each to make up for the lost "memory".
Aside from older hardware that clearly needs it because of sheer lack of RAM, is there even any reason to recommend/enable swap by default anymore? Modern machines come standard with around 4GB of insanely fast RAM - isn't that enough?
What makes this special is not just that it has a cache. Every HDD out there has a cache. This puppy has a "cache" 100x what current drives have.
I think it's silly. The NAND will wear out really quick, and there doesn't seem to be much of a performance boost over a fast WD Black. Actually, there's no performance boost - but it is a smaller form factor that fits in laptops.
I remember a few years back a (Japanese?) company paired 1-2GB of cache with a 5400RPM HDD. They completely maxed out SATA1's transfer speeds - faster than the fastest WD Black released today. (The 2TB one)
If we had HDDs with 2GB of cache rather than 32-64MB, our drives would be a lot quicker. Quite possibly comparable to SSDs in sequential read/write, and a bit higher than our current HDDs in IOPS. I'd take one simply because it wouldn't wear out like NAND does.
I read TFA. The tests were inconclusive. And the drives have to "learn" (most accessed data is written to NAND after a few calls). I like the hybrid concept and anything to boost performance on laptops drives is great. That being said, I would rather see the SSD used like ROM for the OS for FAST boot-up times. Surprised they didn't do it that way.
Don't explain computers to laymen. Simpler to explain sex to a virgin. -- Robert A. Heinlein
But what good is cache if it isn't persistent? The OS already has a perfectly fine read cache. It's the write cache that is the problem, and a non-persistent write cache of multi-gigabytes is pretty scary if you suddenly lose power. You could wipe out an entire file system that way.