Pure computing power isn’t what it used to be. In fact, those who seek it without taking energy consumption into account often end up with an overwrought system, unnecessary costs, and a track record of harming the environment. This is why SAP has defined a benchmark that aids in selecting servers to run its applications by correlating performance and energy use.
Fujitsu was one of the partner companies involved in the development of the SAP Server Power Standard Application Benchmark. Its PRIMERGY RX300 S6 server was the first configuration to undergo the corresponding assessment, the results of which are now available.
The key component of the benchmark is watts per 1,000 SAPS (watts/kSAPS). With SAPS (SAP Application Performance Standard) representing performance, a lower number of watts/kSAPS means higher energy efficiency, as less energy is consumed in handling a certain load. According to the benchmark, the Fujitsu server consumes an average of 18.3 watts in generating 1,000 SAPS of performance.
Here is the full SAP Server Power Standard Application Benchmark certified for Fujitsu’s PRIMERGY server on February 14, 2011:
- Performance efficiency rating: 18.3 watts/kSAPS
- Average performance (all load levels): 11,810 SAPS
- Average ambient temperature: 20.6° C (69.08° F)
- SAP Business Suite software: SAP ERP 6.0 (EHP 4)
- Relational database management system (RDBMS): Microsoft SQL Server 2008 Enterprise Edition
- Operating system: Windows 2008 R2 Datacenter x64
- Main server: Fujitsu PRIMERGY RX300 S6 (two Intel Xeon X5675 processors running at 3.06GHz with 12 cores, 24 threads, 72GB RAM, and a CSCI Gold PSU)
Next page: Measuring different load levels
Measuring Different Load Levels
SAP’s benchmark evaluates performance and energy consumption based on nine defined load levels. Another benchmark SAP uses for its Sales and Distribution component is also applied here. At the 100% level, 4,700 users ran through the SD Benchmark; this is the highest number of users for which the system’s response time is faster than one second. The different load levels have been assigned the following percentages of the maximum number of users:
50%, 100%, 65%, 20%, active idle (one user per SAP instance), 30%, 80%, 40%, 10%
The test is constant and takes around four hours, with the peak interval of each phase leading into the next level. The benchmark determines the respective watts/kSAPS for each load level during these intervals.
In the process, the SAP Server Power Benchmark limits its assessments to the server in question; as such, it factors in local storage units while excluding external media. The benchmark generally incorporates all components installed within the server. Meanwhile, SAP is also currently working on the SAP System Power Benchmark, which will soon provide a means of evaluating the energy efficiency of entire SAP systems.
Next page: The cost of active idle and our summary
The Cost of Active Idle
The graphic above charts the server’s consumption of electricity throughout the benchmark test. The plateaus represent the different load levels; between them, users enter or exit the system to raise or reduce the load.
In the curve below, three phases are apparent. At the beginning, the active idle period is hardly taxing the system, leaving its absolute energy consumption at a low 133 watts.
During the second phase, the curve rises gently but constantly; the load is between 10% and 65%, representing normal system operations. While the system’s energy consumption increases more slowly than its performance in this period, the rate then begins to increase, reaching its peak at 337 watts.
As a rule, the system automatically adjusts to each load level.
The server’s energy efficiency is low during active idle; despite hardly being asked to handle any tasks, it still consumes 40% of the amount of energy required at maximum load. At 65% load, however, the balance between performance and consumption is essentially ideal: The system runs efficiently and the factor between the two variables remains fairly constant. The additional computing power needed to deal with higher loads comes at a cost of ever-increasing energy requirements.
This 65% level can be achieved using intelligent load-distribution concepts in normal operations. Fujitsu’s own FlexFrame for SAP, for example, allocates physical and virtual processing power to applications to support flexibility and efficiency.