The tests were performed on servers configured with 2, 4, 6 and 8 gigabytes of main memory at various transaction processing load levels. The results show that for certain configurations and at certain load levels the Intel Xeon based server was 2.4 to 11.7 percent more power efficient while in other cases the AMD Opteron based server was 9.2 to 23.1 percent more power efficient. In addition, when the systems were idle and waiting for transactions to process, the AMD server was 30.4 to 53.1 percent more power efficient.
Power consumption while the servers are idle is particularly significant since many servers spend most of their time waiting for work. A November 16, 2006 press release(1) from IBM quotes a report by the Robert Frances Group(2) which states that on average servers in datacenters are idle 80 to 85 percent of the time.
Other observations that can be made from the test results include: 1) Larger memory configurations deliver both higher throughput and better power efficiency, 2) Intel's power efficiency advantages decrease as memory size increases, 3) AMD's power efficiency advantages increase as memory size increases, 4) For primarily calculation type workloads, the Xeon delivers 8.0 to 14.0 percent higher peak throughput, and 5) For primarily disk I/O intensive workloads the Opteron delivers 11.3 to 19.4 percent higher peak throughput.
These test results were collected by Neal Nelson's second generation Server Power Efficiency Benchmark. This test is a client server benchmark where world wide web transactions are processed against a server configured with Novell's SUSE Linux Enterprise Server, the Apache2 web server software and the MySQL relational database. The benchmark subjects a server to various user loads, reports the power consumed at each load level and provides meaningful comparisons of server power usage.
Read Neal Nelson benchmark Lab's news article and here's the full PDF report.
Comments
Post a Comment