Delivering performance results with best practices on SLES 11SP2
Bill_Buros 270000QDEP Visits (2010)
By: Bill Buros.
In November 2012, IBM launched a new POWER7+ enabled IBM Flex System p260 compute node. For the marketing and system details, see the IBM PureFlex Systems page. Here on the technical community, we wanted to take a minute to reference the "formal benchmark publishes" that are run and submitted to organizations like SPEC.org. Performance proof points like published benchmarks are a great way to keep score of performance improvements, not only with the hardware systems, but as in our case, the software stacks and enabling software pieces that are required for the best performance on any system.
For POWER systems, our performance teams focus on compilers, Java, and toolchain libraries to best optimize the performance. Best practice guides like the recently announced IBM Redbook on POWER7 and POWER7+ Optimization and Tuning Guide are generally based on extensive work in understanding workloads like the SPEC workloads, and also of course working directly with many customer applications.
For this announce cycle of this new compute node, we took advantage of the latest SUSE Linux Enterprise 11 Service Pack 2 operating system, and published performance benchmarks using the IBM XL Compilers, the IBM Advance Toolchain, and IBM Java.
With each performance benchmark, there are straight-forward ideas, flags, hints, and practical examples hidden in the benchmark disclosure information which are great examples of the process of the optimizing of applications and workloads. These results demonstrate to our Power customers that Power systems continue to deliver leading performance, and the technical means to achieve these results is simple to take advantage of.
In the coming weeks, we'll work to highlight some of these basic techniques, ideas, and tuning tips from the actual runs of each benchmark.