AIX > Administrator > Performance

It's All Relative

Some rPerf Values

Table 1 shows rPerf results for the IBM eServer p5 595 server from IBM's report, "IBM eServer p5, pSeries, OpenPower* and IBM RS/6000* Performance Report," Nov. 18, 2004.

Notice that as the amount of hardware is doubled, the resulting rPerf value is something less than double. Gene Amdahl, mathematically described this occurrence by what's known as Amdahl's Law. In simple terms, this means that as more processors are added to a symmetric multiprocessor (SMP), the last processor added is less efficient than the previous one. Hardware vendors work hard on minimizing these effects through effective use of larger caches, faster processors, etc.

Now let's examine some of the components that can induce large variabilities in predicted performance by using the rPerf benchmark results.


Changes in processor architecture can introduce unpredictable performance results. For example, the POWER3* systems were designed for and possessed relatively good floating-point performance for their time. However, their OLTP-type performance was typically less than the performance of the RS64-IV processor-based systems.

According to Terry Brennan and Bret Olszewski's study, relative performance comparisons between the POWER4+* processor-based p650-6M2 and the RS64-IV-based p660-6M1 showed high variability across a number of workloads. Their experience showed that it's particularly difficult to accurately reflect differences across microprocessor families with a single performance metric.

A more recent reference would be the POWER5* processor. Specifically designed as a simultaneous multithreaded processor (parallel pipelines to allow for two executing threads down each processor core), this processor also requires AIX 5.3 to effectively exploit all of its features. Therefore, comparing benchmark results from this lineage of processors to any other lineage could yield large scatters in predicted performance.

Memory Subsystem

Benchmark results show that appropriately fitted memory subsystems are almost as important as the CPU to overall system performance. Key attributes of the memory subsystem include capacity, latency and bandwidth. Therefore, benchmark results tend to occur with the largest memory configurations possible.

I/O Subsystems

Some workloads require substantial I/O bandwidth or minimal I/O latency to reach CPU saturation (utilization levels approach 100 percent). These workloads require the systems under test to be sufficiently configured with I/O drawers, adapters and disks, such that the I/O isn't the bottleneck. If sufficient I/O bandwidth (minimal latency) can't be supplied on a system, measured performance will be limited by the transaction speed of the I/O subsystem.

Tom Farwell is a technical editor for IBM Systems Magazine, Open Systems edition. He can be reached through

comments powered by Disqus



2019 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.

Achieving a Resilient Data Center

Implement these techniques to improve data-center resiliency.


AIO: The Fast Path to Great Performance

AIX Enhancements -- Workload Partitioning

The most exciting POWER6 enhancement, live partition mobility, allows one to migrate a running LPAR to another physical box and is designed to move running partitions from one POWER6 processor-based server to another without any application downtime whatsoever.

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store
IBMi News Sign Up Today! Past News Letters