Big Blue Runs the Power 740 Through the SAP Data Warehousing Paces
November 29, 2010 Timothy Prickett Morgan
Now that the entry and midrange Power7-based servers are out there, IBM is releasing a smattering of benchmark tests on the boxes to show how they stand up on real-world workloads. The latest is on a data warehousing benchmark test based on code from SAP that Big Blue has been favoring in recent years.
I have a sneaking suspicion that this is not just to highlight the very good SQL processing of the DB2 for i database, but also because no one else runs this benchmark test and therefore IBM can avoid odious comparisons.
As you know, SAP’s ERP software does indeed run on the OS/400 and i operating systems. IBM has run various SAP data warehousing benchmark tests on iSeries, System i, and Power Systems boxes, including the obsolete Business Information Warehouse (BW) test and the more current BI Data Mart test. Up until a year ago, IBM’s Power5+ and Power6 platforms running the i5/OS V5R4 and i 6.1 operating systems, respectively, were the only boxes using this test. And only a handful of machines were tested. IBM then moved on to another SAP data warehousing benchmark with the Power7 lineup, called the Business Intelligence-Mixed Load, or BI-MXL, test.
In the spring, IBM tested the Power 750 server to show how it compared to prior machines, and in the summer, ahead of the rollout of the entry and midrange Power7 servers, IBM ran the same test on a modestly configured Power 740 server. Here’s how it stacked up:
As you can see, the Power 740 was configured with a single four-core Power7 processor running at 3.3 GHz and 64 GB of main memory. It was programmed to hit a database of 300 million records, and running at 99 percent of CPU capacity, the Power 740 was able to process 151,426 query navigation steps per hour using the i 7.1 operating system, its integrated database, and SAP’s NetWeaver 7.0 middleware. If you compare that to IBM’s Commercial Workload Performance (CPW) relative performance benchmarks, it looks like the Power 740 needs more CPWs per unit of SAP database work than the Power 750 that was tested earlier in the year. This is not explained in IBM’s benchmark reports, but it may have something to do with the fact that the memory in the box was only 64 GB instead of the 128 GB used with the Power 750. (Then again, the Power 550 had 64 GB and had a better ratio of query navigation steps per hour to CPWs than the Power 750 or the Power 740. Go figure.) If I had to guess, I would say that the SAP workload is a little more sensitive to clock speed and a little less sensitive to the extra cores and threads. The important thing to note is that an entry Power 740 can do the work of an entry Power 570 from a generation ago.
The other thing to note is that IBM has not yet revealed its tests on the Power 720 machine, which is the workhorse of the IBM i customer base. This is peculiar, and something that should be rectified.