8 socket benchmarking revisited

The fight for the best 8 socket benchmark in SAP is in it´s second round. In the left corner … HP “the inventing garage” ProLiant DL785 G5 with an fighting weight of 72 kg or 160 pounds. The system presents:

HP ProLiant DL785 G5, 8 processors / 32 cores / 32 threads, Quad-Core AMD Opteron Processor 8384, 2.7 GHz, 128 KB L1 cache and 512 KB L2 cache per core, 6 MB L3 cache per processor, 128 GB main memory

It runs SuSE Linux Enterprise Server 10 and Oracle 10g. In the right corner … Sun “The revenge of Andy” Sun Fire X4600M2 with an fighting weight of 40kg or 88 pounds. The Sun system is equiped with:

Sun Fire X4600M2, 8 processors / 32 cores / 32 threads, Quad-Core AMD Opteron Processor 8384, 2.7 GHz, 128 KB L1 cache and 512 KB L2 cache per core, 6 MB L3 cache per processor, 128 GB main memory

The system is powered by Solaris 10 and MaxDB 7.6 After fighting the following results were gathered from the fighting judges: Sun X4600M2 with 39,270 SAPS and 7,825 SD users (Certificate: 2008070). The HP was able to gather 35,400 SAPS and 7,010 SD users. (Certificate: 2008064). The winner by a clear margin: Sun “The revenge of Andy” X4600M2! It should be noted that the Sun system already used the extra soft boxing gloves called “unicode” , whereas HP used the old and outdated boxing gloves “non-unicode”. There is already an discussion raging about the usefulnes of results based on those old gloves. Albeit there deliver more punch you can´t use it in more and more fights in the real world. Okay, fun aside: This benchmark result is interesting, albeit you can´t use this result to compare both machines. In the last benchmark the system HP system used Windows Server 2003 Enterprise Edition and SQL Server 2008. The HP system just had 88.24 percent the performance of the Sun System. In this benchmark the HP system had 90,15 percent of the Sun performance. Two different operating systems, two different databases … but a similar difference? It´s just a speculation … but maybe the HP implementation of 8 sockets is vastly less efficent. Why vastly? You have to remember, that unicode costs you 15% performance. It would be nice to see a result for the ProLiant DL785 G5 with 6.0 unicode, Solaris 10 and MaxDB to get a clearer view to the issue. Update: I digged down into this benchmark. Albeit i have still no hard proof for my thesis, the operating systems or the database seem to be just minor factors to the SAP benchmark. I summarized my findings in “Sun X4600M2 vs HP ProLiant DL785 G5 revisited”.