Here is some intelligence about that IBM claim that its SVC is faster than solid-state storage, courtesy of two SPC benchmark results.
Woody Hutsell, a VP at solid-state disk supplier Texas Memory Systems said: "IBM currently holds two SPC-1 results that are faster than those posted by Texas Memory Systems. When we published results in March 2004 with our 2Gbit/s interfaces, we had the best number."
"We have been looking to repeat testing with our 4Gbit/s controllers to demonstrate that we continue to have the fastest storage. Our challenge right now is getting a server configuration capable of driving the RamSan to show off our performance which we are positive is faster than the IBM RAID systems."
"While we do not currently have the top SPC-1 marks for performance, we continue to lead the pack in SPC-1 Cost per IOPS. It is worth looking further into the SPC results posted by IBM and Texas Memory Systems, which would reveal the following:
1. For IBM's SVC test. a 42-CPU P5 595 server with 32GB of memory is used with 32 Fibre Channel links. If you look on the TPC website you can get an idea of the cost of this server. A result from this year shows that a slightly faster 2.3 GHz processor (the SPC test mentions 1.9GHz CPUs are used) has a per CPU license activation fee of $30,000. With just this portion of the cost the test server is well into the million dollar range. The DS8300 test uses 32 of the 1.9 GHz P595 processors. Texas Memory Systems does not have the resources available to devote a million dollar server to the testing. Note that simple white boxes were used for our results, which in 2004 cost roughly $20,000 for 8 systems.
2. The second thing to pay attention to in the test results is the response time of the tested storage configurations. This information is available on page 4 of the SPC-1 executive summary reports. For the SVC and the DS8300, you can clearly see that the response time with IOPS curve has an inflection point indicating that the array has been saturated. With a load of ~110,000 IOPS (near the maximum reported by the RamSan testing in 2004) the SVC has a response time of ~7 ms and the DS8300 has a response time of 5.5 ms. The response time of both IBM arrays then quickly climb to 15 ms. Now look at the RamSan-320 response time with IOPS load: at the 100 percent load the response time is 0.73 ms - dramatically lower, and, furthermore, there is no inflection point on the graph - the array has not been saturated!
3. We guarantee that our system will provide the highest IOPS performance for a random small block workload then any other storage device under our application performance guarantee program and will fully refund the purchase price of a RamSan if any storage equipment in the data center out performs it."
It seems blindingly obvious to me that, if the disk resources on the IBM test configuration were replaced with solid-state drives and equivalent speed connection links, the solid-state configuration would simply blow away the HDD one.
Anyone care to stump up a million bucks to prove it?