The Quantifiable Benefits of SSDs

It's not my site, Eremeeff. If you want something, go ask the EDN staff by using the Nav bar. You'll find "contact us" under the button marked "More".

Can i get a one small picture from your site? Eremeeff

Tired of Hype: I will not defend a particular vendor's drive. I also do not test drives. I report what experts say when they stand up in front of an audience. If STEC drives get 11x, then that's what they get. If you look at pricing, Intel's not getting 100x pricing over spinning disks. I think you exaggerate. It's more like 4-5x. And Intel's supposed to have the fastest SSDs in the business. When Sumeet Bansal reported getting 1200% performance improvement in Writes for wine.com, that's what he reported. Didn't sound like hype to me. I've already agreed with you that the TPC-C benchmarks should be high on vendors' proof-point list. However, just because a raw SSD gets 100x the IOPS, I do not expect a system to get 100x performance improvement in benchmarks like TPC-C. For example, Windows systems get something like a 30-40% improvement in SYSmark performance with an SSD, because once the drive becomes less of a limiting factor, other factors slow performance to a new plateau. This is simple Amdahl's law. Speeding up one element in a system doesn't improve system performance by an equal amount because each element's performance only contributes a fraction to overall system performance. Maybe a 30-40% improvement in a Windows system or a 10-12x improvement in a server isn't enough for you. So be it. A lot of system designers kill themselves for 5-10% and SSDs seem to provide a lot more with a plug-in component replacement. If that's not a good tradeoff for your situation, then don't buy it.

Steve, I don't think I suggested that SSD vendors "deserve no coverage". But you said: "As I?ve been discussing in my most recent posts, SSDs deliver 100-200x the IOPS (I/O operations per second) performance of hard disks." That's not true anywhere in the world outside of the vendor spec sheet and benchMARKETING claims. For example, when IBM published the world's first audited, verifyable benchmark of an SSD (the $23,000 STEC ZeusIOPS) in June, they they found (a) that it only produced 10% of the IOPS claimed by the vendor, and (b) that it could only do about 11x the number of IOPS of a 15Krpm spinning disk. Not 200x, not 100x...11x. So, it COSTS 100x more than spinning disk, but only delivers 11x the performance. That's a big hype/reality ratio. (See the SPC1-C/E results at wwwstorageperformanceorg Maybe instead of complaining about people who are tired of the hype and want to see real performance data, you should examine the claims vendors make in light of independent evidence, and stop feeding the hype-cycle.

Tania: These blog pages are generated dynamically by EDN's database. You're not even allowed to put a hyperlink in the comments. Text only. If you see a virus warning, it's not coming from here. Some other parallel stream, perhaps?

Hi, Onload of page my antivirus put alert, check pls. Thanks Tania

Tired of hype, I agree with you that benchmarking SSDs on the TPC-C benchmark is an important step for SSD vendors to take. Your caustic remarks suggest that SSDs deserve no coverage until the vendors comply with your wishes. As a blogger, I cannot agree. I attended a conference where something interesting was said and I conveyed that information as transparently as I could to thousands who could not attend. Personally, I think this is a terrific use of a blog. If you don't like this sort of thing, then why not ignore it? As for free advertising, it's the same endorsement-free advertising that any product written up in EDN gets. I do not think a case study is a wild claim. The wild claim would be that every integrator that uses an SSD would get the same results. Such a claim was not made by the speaker nor suggested in the post.

While wine.com teams up with Fusion-IO for some free advertising, I note that Fusion-IO and all of the other SSD manufacturers have steadfactly refused to subject themselves to third-party-audited TPC-C or TPC-E transaction processing benchmarks that would prove out these claims. In the only audited benchmark result yet published for SSD, a system based on conventional spinning disks was twice as fast as the SSD system (in queries-per-hour) and cost half as much. See for yourself at www.tpc.org. Also, note how neither this article (nor the case-studies released by the vendor) tells us anything at all about what the wine.com storage infrastructure looked like before the SSD upgrade. The TPC-C has been the "gold standard" performance benchmark of database performance and >>cost/performance<< for over 20 yrs. Every major database and systems vendor uses it and there have been many hundreds of published and audited results. With these wild claims, why are the SSD vendors not playing?