MS Researchers Call Moving Server Storage To SSDs a Bad Idea 292
An anonymous reader writes "As an IT administrator did you ever think of replacing disks by SSDs? Or using SSDs as an intermediate caching layer? A recent paper by Microsoft researchers provides detailed cost/benefit analysis for several real workloads. The conclusion is that, for a range of typical enterprise workloads, using SSDs makes no sense in the short to medium future. Their price needs to decrease by 3-3000 times for them to make sense. Note that this paper has nothing to do with laptop workloads, for which SSDs probably make more sense (due to SSDs' ruggedness)."
Re:they already cost less per gig than some SAS dr (Score:3, Interesting)
Of course, SAS drives are also often too expensive to survive a purely cost/benefit driven analysis. For many real-world loads you're better off adding more spindles which can give you similar iops per dollar but with the added benefit of vastly more storage space.
There's a lot of snake oil and very little quality analysis in enterprise storage these days, so it's good to see at least some do attempt to do actual real-world cost/benefit calculations before jumping onto the marketing train.
Re:Read the Paper (Score:1, Interesting)
One of the notable parts of the paper is they use 10k-100k write cycles as the maximum for the flash media(page 5, Reliability and fault-tolerance). Change that to the 1 million most manufactures are reporting and the numbers would be significantly different. Something else on this bit, is that the papers they reference for higher failure rates of DISK drives are 2-4 times higher so even using those numbers the failure rate microsoft used(10-100 times) seem to be skewed. And the authors appear to completely ignore this evidence for the evaluation of the disk drives.
Then change the prices to something more reasonable like $725 for 250gb, a fresh number from newegg.com for a OCZ ssd. Instead of the $730 for a 30gb drive and the numbers change radically again.
I couldn't find much information in English about the memoright drive(i spent maybe 15minutes on this total) they used but its appears on par with the spec for most other MLC based flash drives on the market currently.
I would guess they drive performs about 6-10 times better than their paper states.
Just another bit of FUD from m$.
Re:Read the Paper (Score:4, Interesting)
They tested precisely one brand/model SSD, as far as I can tell from the paper. It did 351 random writes per second, which is pitiful (but probably typical). Intel claims 3300 random writes per second for the X25-E.
Re:This is probably a reaction to Sun's L2ARC (Score:2, Interesting)
Re:Read the Paper (Score:3, Interesting)
I would not call an OCZ Vertex drive enterprise level, but you can already see that changes in the controllers can make massive differences. Just compare the random write speeds in this article with those in the latest reviews of Anandtech. Once new controllers start coming out and compete (both in performance and price), the landscape will be entirely rewritten by SSD.
And although the Memoright is still very expensive, in general you see a massive change in price for SSD's. Currently they seem temporarily *more* expensive, but I presume that's because of strong demand. A price reduction of 10x (or an order of magnitude, in the authors words) when you start off with ~770 dollar for 32 GB is completely possible.
Things like boot times (for those people needing restarts and high availability) and latency, both highly in favor for SSD are missing entirely from the article.
I don't know if this is FUD, but I would definitely not call it accurate - the reviews of Anandtech seem much more precise and much more valid. And Anand reaches an entirely different conclusion.
Re:Not every tool is right for every application?! (Score:2, Interesting)
Plus with the wear levelling feature, the useful life of and SSD hammer seems like it would be much more reliable over a spinning disc hammer.
How does wear leveling fair when most of the content on the drive isn't being changed ever, while some files are changed quite frequently? If 60% of the drive's contents don't change (which, if you have a computer which you're using and not frequently installing / uninstalling programs), doesn't that mean that the 40% of the rest of the drive is being worn evenly, but that other 60% still has a lot more life left?
Re:Not every tool is right for every application?! (Score:5, Interesting)
I love that term "Enterprise storage".
A hard drive is a hard drive is a hard drive. It doesn't matter whether you pay $79 or $799 for a drive, it is just as likely to crash, burn and lose all your data as any other. The sole difference between "consumer" and "enterprise" drives lies in the firmware. It might have more aggressive queue deadlines, or be configured to "fail" as soon as a single defect is identified (even though it can remap them with spare sectors). In many cases they are 100% identical.
It's all just markup, and in some rare cases you might get an extra year or two on the warranty. I'd much rather buy ten cheap drives and have a bunch of spares, than buy the "enterprise" model and have it die just as catastrophically.
There's no shortage of FUD threatening that if you use a cheap drive in a server, your wife will cheat on you with a Seagate engineer, your first-born child with "go gay", coworkers will laugh at you and call you "Cheapy McCheaperson" behind your back, and Larry Ellison will reach down from his bearded throne and slap you across the face with a greased-up midget.
If an enterprise SSD comes with an enterprise-class warranty that justifies the cost, great! If not then you're a sucker for buying it, as the vendor laughs all the way to the bank.
Re:Read the Paper (Score:1, Interesting)
Serious users are already moving to products like the FusionIO where the FlashRAM is mapped directly onto the PCIe bus instead of though legacy interfaces like SATA.
They already ship products that deliver > 1GB/s sustained and over 100,000 IOPS on a single card. No production mechanical drive array even comes close.