Understanding And Interpreting Specpower_Ssj2008 Results - Compaq BL10e - HP ProLiant - 512 MB RAM Update Manual

ISS Technology Update

Understanding and interpreting SPECpower_ssj2008 results

Server benchmarks have traditionally focused almost entirely on server performance. As server power consumption and
efficiency have become increasingly important issues, the need for objective and reproducible measures of server performance
relative to power consumption has also become important. In late 2007, the Standard Performance Evaluation Corporation
(SPEC) introduced the first SPECpower benchmark (SPECpower_ssj2008) to address this emerging need. SPEC is a non-profit
corporation formed to establish, maintain and endorse a standardized set of benchmarks for high-performance computers.
SPEC is supported by its member organizations which includes HP.
Structure of the SPECpower_ssj2008
The SPECpower_ssj2008 benchmark measures server performance in relation to its power consumption. The benchmark allows
the use of various accepted power analyzers to measure the server's power draw. The benchmark harness is a program that
receives power data electronically without human intervention. The program launches the workload and collects power and
thermal sensor measurements. It then post-processes and reports performance, thermal, and power data.
SPECpower_ssj2008 workload
SPECpower_ssj2008 uses a server side Java (ssj) workload to generate work for the server and to measure the compute
performance. This workload met SPEC's requirements:
 It is scalable and multi-threaded.
 It is portable across a wide range of operating environments.
 It can be run quickly and efficiently at minimal cost.
Using a graduated workload to measure performance
SPECpower_ssj2008 uses a graduated workload to measure server performance and multiple target loads. With this
methodology, the server performance and power consumption are measured at eleven different levels of server utilization for a
fixed period of time (240 seconds). This provides a more complete picture of the server's performance and power consumption
across a range of operating conditions.
For example, the benchmark results for the HP ProLiant DL385 G6 are shown in Table 1-1. The results include the server
performance and power consumption at eleven different levels from zero to one hundred percent load. Zero percent load
represents the active idle state of the server. The result is a performance-to-power ratio calculated for each ten percent increment
of server load.
Table 1-1. Published SPECpower_ssj2008 result - Power performance and power summary for the HP ProLiant DL385 G6
Target Load Actual Load
100%
90%
80%
70%
60%
50%
40%
30%
Server side Java
Operations (ssj_ops)
99.4% 535,814
89.7% 483,541
80.0% 431,269
70.1% 377,831
60.2% 324,171
50.0% 269,338
39.7% 213,708
29.7% 160,021
Average active power
(watts)
260
246
234
220
207
194
181
168
Volume 8, Number 6
Performance to Power Ratio
2,062
1,962
1,847
1,715
1,565
1,386
1,181
953
2