Reader Feedback On Oracle Takes On IBM Power With New Sparc T5 Systems
April 8, 2013 Hey, TPM
I won’t say “Is this a joke?” although it is very tempting.
I enjoy reading your articles, even though I had only a light period with managing any OS/400 (V4R2 days). My focus is AIX and Unix, since 1979 actually.
When you test these systems look closely at the specs for transfer between chips and cores. My experience with AIX and Power is that customers do not like doing the planning required to get the most out of the box. NUMA is “boring,” and planning your workloads to work with it is a headache for some; not doing the planning can be a nightmare.
Question: Has Oracle published its benchmarks? I have not been looking hard, but I did not see any reference to any in your article either.
Something else I am curious about: If Oracle makes special processing hooks for Oracle, are they not required to publish the specs or run the risk of Europe and/or USA agencies starting an antitrust action? (Not my area, but it is a question that crossed my mind.)
The T5, based on what I am told, does not worry me technically–so I appreciate your more “neutral” presentation. I expect them to have lots of surprises when customers try to get all they can out of it. Paraphrase: “16 DIMMs, one for each core–for balance!” When I read that I thought–many sleepless nights and terrible dreams. Glad it is not my box.
It will be interesting to see what will happen as both IBM and Oracle tune up their software to take advantage of specific features in their hardware. I am not sure what the legality is regarding providing access to the specs of those hardware-specific features for accelerating software, such as databases, but it seems to me that you are absolutely allowed to innovate and make software run better. The issue becomes one of when one vendor wants to have access to those features.
I can’t imagine IBM making a big fuss and wanting to use database acceleration features for DB2 when it runs on Sparc-based systems. But DB2 8.2, 9.1, 9.5, 9.7, and 10.1 are supported on Solaris 10 and 11 operating systems running on Sparc T series and Sparc Enterprise M servers, so technically it is possible that IBM may want to do this.
I think it is far more likely that Oracle and IBM will square off with their respective hardware and software stacks and increasingly tune their software to run with their specific hardware. The trouble is that no matter how well IBM has done to improve DB2’s market share among the Unix faithful, Oracle still is the dominant supplier of database software on Unix systems, and that means IBM needs for whatever optimizations Oracle cooks up for Sparc to translate over to Power iron. IBM cannot give Oracle a big performance lead on Sparc iron over Power iron, or it will start losing Oracle database customers who will shift back to Sun hardware.
The question really, then, as far as I can see is: What can IBM do to give DB2 a big performance edge while keeping parity with Oracle on its own database? IBM can add lots of things to DB2 to speed it up, as it demonstrated with the BLU Acceleration feature for DB2 10.5 that IBM previewed last week. BLU Acceleration is analogous to the hybrid columnar data store that Oracle cooked up for its Exadata storage servers, and is only available for reporting and analytics workloads, but should make its way into transaction processing systems at some point. Oracle can’t just give up on Power Systems iron, either, although as Oracle’s behavior with Itanium systems from Hewlett-Packard show, if Oracle thinks it can gain an edge by removing support of a platform, it is perfectly willing to undermine that platform for a longer term gain. There is no question, two years after saying it would stop developing software for Itanium that the Sparc platform is perceived as having a more viable future than the Itanium platform. Neither product is selling at anywhere near the volumes of IBM’s Power Systems, of course.