Sun X4600M2 vs HP ProLiant DL785 G5 revisited

I further digged down into the topic of the seemingly vastly slower HP Proliant 785 G5 when compared with a Sun X4600M2. In some discussions colleagues told me: You have to compare two systems with the same os and same database. This discussion even found it´s way to the german heise.de forum. Okay, that´s correct for a final proof but i don´t expect a HP running the SAP SD benchmark on Solaris and MaxDB, so we will never know it …. So i´ve studied the SD 2-tier benchmark list out of curiosity to find some hints to explain the performance difference between the both benchmark results.

Hypothesis 1: The different database and operating system explains the benchmarking difference

This would be the most obvious explanation. The Sun benchmark used Solaris 10 and MaxDB and the both benchmarks done by HP uses Linux/Oracle respectively MS Windows/MSSQL. But then i found the following benchmarks in the benchmarking list. There was a benchmarking run (SAP Certificate: 2006079) with an HP system on November 1, 2006 performed by HP:

HP ProLiant DL380 G5, 2 processors / 8 cores / 8 threads, Quad-Core Intel Xeon Processor X5355 2.66 GHz, 64 KB L1 cache per core and 4 MB L2 cache per 2 cores, 32 GB main memory

This system running with Windows Server 2003 Enterprise Edition and SQL Server 2005 delivering 8970 SAPS. On February 12, 2007 there was a benchmark (SAP Certificate: 2007028) with a server configured as following:

HP ProLiant DL380 G5, 2 processors / 8 cores / 8 threads, Quad-Core Intel Xeon Processor X5355 2.66 GHz, 64 KB L1 cache per core and 4 MB L2 cache per 2 cores, 32 GB main memory

You see, it´s the same system. This It was operated with SuSE Linux Enterprise Server 10 and Oracle 10g. This system yielded 9000 SAPS. The difference of two identical configured system with different OS and DB was just 0.34 percent. It´s just an educated guess, but the choice of database or operating system just seems to have a minor impact to the SAPS number, at least when comparing Linux+Oracle and Windows+MSSQL.

Hypothesis 2: The MaxDB database from SAP delivers more performance than Oracle 10g or MS SQL

This would be the next obvious conclusion. Sun used a different database than HP. But there is some data in the benchmarking list that suggest something else. On April 19,2008 HP performed a benchmark (SAP certificate: 2008025) on the following configuration:

HP ProLiant DL380 G5, 2 processors / 8 cores / 8 threads, Quad-Core Intel Xeon Processor X5460, 3.16 GHz, 64 KB L1 cache per core and 6 MB L2 cache per 2 cores, 32 GB main memory

This benchmark was performed with MaxDB 7.7 and yielded 11400 SAPS. On October 30, 2007 performed an benchmark (SAP certificate: 2007064) with the same configuration, but used Windows Server 2003 Enterprise Edition and SQL Server 2005. This yielded 12,180 SAPS. It´s yet another educated guess, but the usage of MaxDB seems to slowdown the system. There is no proof for the hypothesis that MaxDB is better than Oracle 10g or MS SQL.

Conclusion

There are two remaining hypotheses. The first one: “Solaris 10 explain the difference between the Sun SAPS number and the HP SAPS number and is vastly better than Windows and Linux in regard of SAP”. I can´t support or dismiss it with numbers, because there is is no SAP benchmark in the list which only differ in the operating system but not in hardware and database. The second one: “The X4600 is vastly better than the HP Proliant DL785 G5 in regard of performance”. As there is no benchmark of both system with the same combination of OS and database i can´t support or dismiss it as well. Anyway … the proof of any of the both both hyptheses would be very good for Sun. Either our OS is vastly better or our system is vastly better. But both would proof: Engineering isn´t futile.