SPEC Members Start on Energy Benchmark for Web Servers
June 16, 2008 Timothy Prickett Morgan
There’s an old adage: You can’t manage what you don’t measure. As most data center managers and server and storage arrays makers have been well aware of for years, we have some pretty serious power and cooling issues affecting IT operations. (As I write this, the 95-degree temperatures in New York City have me worrying about keeping IT Jungle’s Linux-Windows cluster cool once again drives the point on home.) While we have many different benchmarks for determining performance of computers, there are few workloads that have been tweaked to provide a standard means of gauging power draw while benchmarks are running.
This is a problem, and the Standard Performance Evaluation Corporation, one of the key providers of benchmarks used by the industry, says that it is now working on its second power-related benchmark test, this time around, one that measure the power efficiency of Web serving.
“Power used by data centers is expected to double in the next five years and Web servers will likely play a prominent role in that growth,” explains Rema Hariharan, the chairman of the SPECweb committee that is putting together the new test and a former technical staff member of AT&T, a performance engineer at Sun Microsystems, and now hailing from Advanced Micro Devices. “A SPEC benchmark could help abate this trend and establish more power-efficient practices.”
In January, as we previously reported, SPEC put out its first benchmark that gauges the performance and power consumption of servers, called SPECpower_ssj2008. While the code used in this test is different from others, it beats some resemblance to the SPECjbb2005 Java server benchmark test. Importantly, the SPECpower_ssj2008 test not only gives us a standardized test to measure power consumption at various performance levels, but sets out a methodology by which other benchmark tests can be tweaked to give them a consistent set of power consumption metrics. This will probably guide how the SPECweb committee develops a power-sensitive and power-measuring version of the SPECweb2005 test.
AMD, Fujitsu-Siemens, Hewlett-Packard, IBM, Intel, and Sun are all on the SPECweb committee that is coming up with the Web power rating benchmark, and they are all soliciting input from corporations, universities, vendors, and any other interested parties. The new SPECweb power test is expected to be launched at the end of 2008 or in early 2008.
SPEC has also announced that it has updated the SPEC CPU2006 suite of integer and floating point performance tests, which are perhaps the best known of the SPEC family of benchmarks. With CPU2006 V1.1, SPEC has implemented more portable code, which will allow the benchmark test to be moved from machine to machine in a family of products and across families of compatible products without having to recompile the benchmark code in each individual machine. The new suite also is easier to set up and run on larger systems, particularly those with lots of processors and cores. (The benchmark has to reach a steady state of operation before the test results can be gathered from the system, and this process of ramping up the workload used to take a lot more time.) Companies using the V1.1 code to test their boxes can compare them to prior V1.0 release results; the changes in the code are about usability of the test code but have no effect on performance results, according to SPEC.