Slashdot Mirror
Intel Replaces Consumer SSD Line, Nixes SLC-SSD
Lucas123 writes "Intel today launched a line of consumer solid state drives that replaces the industry's best selling X25-M line. The new 320 series SSD doubles the top capacity over the X25-M drives to 600GB, doubles sequential write speeds, and drops the price as much as 30% or $100 on some models. Intel also revealed its consumer SSDs have been outselling its enterprise-class SSDs in data centers, so it plans to drop its series of single-level cell NAND flash SSDs and create a new series of SSDs based on multi-level cell NAND for servers and storage arrays. Unlike its last SSD launch, which saw Intel use Marvell's controller, the company said it stuck with its own processing technology with this series."
The 320 series isn't quite as impressive over the X25-M G2 series as I had originally hoped, so will likely be quite some time before I bother replacing the current one (and move that into the laptop instead).
Still, an update has been due for a long time now the X25-M G2 is ancient in SSD terms. Just hope the new controller is as reliable as the Intel one found in the old drives.
MLC the only option on a server? For high-transaction databases, I don't see how it will work.
I'm not going to run out and replace my $100 2TB external backup with one of these any time soon. However, I've been tempted to snag a small 40 gig model and use that as my OS drive, and use my existing internal 1TB HDD for the actual data. I think the article is right, in that the price per gig needs to hit $1 before you start seeing acceptance for mass storage solutions from consumers. 95% of users can't tell the difference between a 5600 RPM HDD and a 10,000 RPM one, so they won't care about SSD speeds that much either.
Occasionally living proof of the Ballmer peak.
Seriously. Any sort of enterprise-level should be swearing off these things as a storage medium then. Well, maybe for a boot drive. But anything with massive amount of writes should be kept as far away from an MLC drive as possible.
Chas - The one, the only.
THANK GOD!!!
It is a bit behind the times with no Sata 3 (6 GBps) support.
Ian Ameline
Why? If the MLC cells are both fast and reliable, why does that matter? If I understand this correctly, MLCs would be the equivalent of clusters on an HDD. If any bit of that data within that cluster needs to be changed, its entire contents will be all read, and re-written back to another cluster. The same process occurs on an MLC.
Life is not for the lazy.
Because SLCs survive for two orders of magnitude more writes than MLCs.
Looks like like Intel has scrapped the "power safe write cache" that was slated for the next generation of drives.
Also, each cell write is much faster (because it can be "sloppier" with only two states per cell), which greatly affects random write speeds even if the speeds are the same for sequential writes.
And random writes is often a bottleneck in master databases.
Because SLCs survive for two orders of magnitude more writes than MLCs.
I don't work with this sort of stuff, but does that matter? If MLCs have other advantages, then what the problem with chucking them out and replacing them when they wear out?
So, anyone cares to make a forecast as to when SSDs will overtake HDs even for large Tb units in terms of price+perf ?
Non-Linux Penguins ?
Place your webbrowser caches onto it
Very bad idea. That's random writes, where SSDs are typically far slower than normal hard disks.
In particular, many users with SSDs complained about freezes for several seconds at a time when using Firefox, and it turned out that all the updates to the history and cache indexes were the root of the problem. Moving it to a HDD drive, and the problem was gone. No, this was not a bug in Firefox, but a side effect of how SSD operates. IIRC, the Firefox developers still added a workaround, which to some degree alleviated the problem at the expense of a risk of the history and cache not necessarily being in a consistent and up-to-date state if the browser crashes.
Because two orders of magnitude is the difference in price between a Honda Civic and a Lamborghini Gallardo.
More to the point, it's the difference in life between one month and 8 years.
Place your webbrowser caches onto it
Very bad idea. That's random writes, where SSDs are typically far slower than normal hard disks.
That's a problem with some controllers (Marvell's Da Vinci controller and whatever the hell WD uses in their ones, for example) that don't handle garbage collection well. after being in use for some time, the write speeds and IOPS goes all over the place. It's not a problem with better controllers, like Sandforce's stuff.
the SSD market still has a bunch of garbage in the mix and you need to research stuff to avoid getting burned.
upon the advice of my lawyer, i have no sig at this time
Because two orders of magnitude is the difference in price between a Honda Civic and a Lamborghini Gallardo.
It's a lot closer between the max speed of a Honda Civic (117 MPH) and something that can cross the Atlantic in 29 minutes.
No, I'm pretty sure this Civic could cross the Atlantic in 29 minutes... ;)
There's no comparison between the 5,600-10,000 RPM gap and the HDD-SSD gap.
I took the plunge last year and installed X-25M drives in my desktop and laptop as OS drives, with secondary drives for user data. The difference is the single greatest performance jump I've ever experienced in 30 years of upgrading, going even back to the days of replacing clock generators on mainboards to overclock 8-bit CPUs by 50 percent.
There is literally a several-orders-of-magnitude difference in the overall speed of the system. If you haven't experienced it, a description of the difference doesn't sound credible, but a multi-drive RAID-0 array of 10k drives doesn't come close to a single SSD in terms of throughput.
I can't go back to non-SSD OS installs now. Systems without an SSD literally seem to crawl, as if stuck in a time warp of some kind. Non-SSD systems seem, frankly, absurdly slow.
STOP . AMERICA . NOW
Thats why they now have eMLC drives... which, if I'm not mistaken, is what the intel drives are. The e in eMLC stands for Enterprise. They can do 3x the number of write/program cycles as the average MLC drives. While still not as good as SLC (or eSLC, which is also out there), it's not as bad as it could be.
But really, for enterprise level storage, you should still stick with spindle based storage, and use MLC drives for read cache, and a mirrored pair of eMLC, or SLC drives for write cache. Or, in ZFS parlance, use MLC drives as cache, and eMLC/SLC as mirrored log drives in a zpool.
OMG... I have a sig?
No. Two orders of magnitude is 100x. Good SLC vs good MLC is 10x, only a single order of magnitute longer lasting.
http://www.anandtech.com/show/2614/4
What you forget is MLC is about 2x cheaper than SLC, so you can get 2x the space for the same price. With wear leveling, extra space is extra lifespan, so MLC dies 5x faster than SLC.
What does that mean for you? I put my money (job) where my mouth is. Our reasonably high traffic OLTP database server uses Intel SSDs as filesystem-level write cache. We get an average write level of 10MB/sec. The minimum expected lifespan of the drive is 2 petabytes. That means we likely have SIX YEARS before the cells start to become unwritable. At that point, no data will be lost: the drive will report the write failures to the OS and store to cells that haven't become unwritable yet, and you will be able to continue operating for the next few months while you get a replacement drive.
If it's in you sig, it's in your post.
Comment removed based on user account deletion
Or decades and years. After a year in a database server at 195 writes/sec (after merge), 1MB/s average, 24x7, which would easily saturate 4 disks at 10k RPM, the X25-M 80G with MLC cells says it's still fine. Smart attribute 233, Media_Wearout_Indicator, is at 95% of new. 19 more years should be enough, I doubt any other component in the server would last this long.
SLC, MLC, it doesn't matter that much, SSDs are still more than capable to replace a whole stack of spinning media in IOPS. For Intels, just use 20MB/s = 1 year for 80G, 40MB/s = 1 year for the 160G version. Mind you, one megabytes per second for 24/7 is a *lot* of INSERT queries, database servers are high IOPS, but low throughput. It's enough for ~50% of the pageviews of Slashdot in our case.
I know that random writes is a problem - which is why I mentioned things like 'enough cache' and 'right database system' to turn those random writes into sequential ones. It's expensive in terms of storage capacity(your DB will have to be bigger), but if MLC is 'enough' cheaper than SLC, you just buy the additional storage. Plus, with modern MLC and wear-leveling you're looking at years at the drive's maximum write speed to start wearing out the cells. If MLC is around an order of magnitude cheaper than SLC, it's cheaper just to replace the drive more often - especially with prices for a given size dropping all the time.
I don't read AC A human right
On the bang for your buck argument, I'll stick with a platter hard drive. I've had laptops for what seems like forever. I carry one in a shoulder tool bag that gets bounced around on a two wheel cart, plopped down on concrete 6-10 times a day, and in 10 years, I think I've lost one drive. Until the price per megabyte comes down close to that of a traditional platter drive, I'll stick with them. The technology is bang on reliable, speed is pretty good, and the "green" drives use less current. The SSD's are still too pricy, the performance "boost" isn't that impressive.
Thanks, my mistake. I looked over the datasheet and product brochure and it made no mention of this. Since they were touting it prior to launch, it seems strange that it is no longer a marketing point. Hopefully the feature won't disappear, as has happened with certain other products after launch.
Doubling lifespan that way requires that you only use half the disk capacity.
I have burned out a Major Name Brand SLC SSD with a high traffic OLTP DB in eight months. I have heard the same from Large Internet Companies which tested these for internal use. There are ongoing independent reliability expert studies in FAST, HOTDEP, other conferences which are uniformly highly skeptical of vendors' claims on SSD lifetime.
If you have not actually tested the drive out to six years service, run an accellerated pilot test unit out ahead of your main prod usage, to give you the canary warning.
Also, each cell write is much faster (because it can be "sloppier" with only two states per cell), which greatly affects random write speeds even if the speeds are the same for sequential writes.
And random writes is often a bottleneck in master databases.
I hear this come up every time even though existing SSDs, both MLC and SLC, already run circles around hard drives for both random read and random write performance.
I have an old SSD in my laptop that can outperform a very expensive SAN array for database workloads -- I've tested it with the same database and the same query side-by-side with the 16-core production database server with a 48-spindle LUN behind it, and my laptop won every time.
Stop quoting stuff that was barely true for some first-generation drives. The latest stuff is faster than a spinning disk at everything, often by orders of magnitude.
Pfft, I think that Civic is probably high-centered on a blade of grass and can't move from that spot.
If you are not allowed to question your government then the government has answered your question.
Please fix your writing style. It is quite grating.
I have burned out a Major Name Brand SLC SSD with a high traffic OLTP DB in eight months.
Why tip-toe around this? Are you talking about Intel or not? If not, it's not really relevant here because this is about Intel, and I think most people agrees that Intel is generally a bit more respected for being a better tested product with a bit more truth behind their numbers. If you ARE talking about Intel, then I think that's pretty important to know.
Are such reports available to the public? I'd love to read over them. If not, what major brands are you referencing? At the very least, was Intel on that list?
Life is not for the lazy.
I hear this come up every time even though existing SSDs, both MLC and SLC, already run circles around hard drives for both random read and random write performance.
That's rather irrelevant when you compare two SSDs, though? Who said anything about HDDs?
(And besides, it's not true. The average speed is far higher for SSDs, but a short-stroked HDD has a much better worst case random write speed than any SDD currently on the market. Yes, really. And for some uses, that's what matters. But again, that's not relevant, because we're talking about SLC vs MLC drives here.)
Doubling lifespan that way requires that you only use half the disk capacity.
Actually, the amount of data that can be written is: the capacity of the device that is used × the number of times it can be written. For example, a CD-R and a DVD-R can both be written only once, but you can write at 10MB/s to a DVD much longer than you can to a CD-R. Using only half the capacity of the DVD-R wouldn't help, and would in fact halve the amount of time you could write 10MB/s to the DVD. A SSD is similar, except that it has multiple write cycles. The way wear leveling works is that writes are are distributed evenly across the medium, so you always use the full capacity of the device.
We use GNU/SunOS.
This is a consumer drive. The "M" in the X25-M stood for "mainstream". As it is replacing the X25-M it is also a consumer drive.
Not a sentence!
georgewilliamherbert does not have any good points as far as I can see. His latter comments are offtopic, complaining about non-Intel flash-based drives in general, when this thread is about Intel changing from SLC to MLC. (Ignoring that, I find his lack of references telling.)
To address his former point, the Intel drives use static wear leveling (as far as I know) so even if the drive is full, the flash is fairly evenly worn. That means you can get more space for the same price which mitigates the shorter lifespan of MLC.
Nowhere do I advocate cheaping out on your components, nor are my remarks aimed at home businesses. I think the rest of your comments about operating costs are a bit misplaced.
If it's in you sig, it's in your post.
One thing to consider is the price per capacity and how that affects performance. You can get Intel MLC based SSDs for about $2.15/GB and SLC based SSDs for about $11.70/GB. That can translate into 4x the number of drives at the same capacity, which is four times the controllers working together in a storage system. Then you can front end that with large write caches and you _MIGHT_ end up coming out ahead in performance.