Elisabeth Stahl on Benchmarking and IT Optimization

Cisco/Oracle Super Saturday TPC-C

with 2 comments

So now they’re trying to put another one over on us. “Super Saturday” for retailers, where you shop till you drop the Saturday before Christmas, has been officially moved to tomorrow, one week early. To give us more sales and more time to shop until we drop. I’m just getting tired of being told that I should shop on Black Friday, Cyber Monday, Super Saturday, or any other time. It’s sort of like the greeting card companies saying let’s make a new holiday to celebrate your dog.

And that’s how I’m feeling lately being told about the latest Oracle and Cisco benchmark claims on the newest TPC-C OLTP benchmark result. Here are the 3 things you need to know:

  • The Benchmark: Oracle didn’t even run this benchmark. And they didn’t even run it on their own hardware.
  • The Performance: Note the special pricey Violin memory arrays that Cisco/Oracle used. And even an IBM result from over a year and a half ago on this benchmark is 70% better performance per core than this result.(1)
  • The Pricing: Cisco and Oracle both take advantage of Super Saturday pricing methods in these results. Cisco shows a 57% “large purchase discount” for hardware. Pricing for Oracle 11g is for the limited standard edition. And very minimal support is included.

I know that both American Greetings and my black lab would really like that idea for the new holiday to celebrate your dog. I already bought my lab a large braided rawhide with red and green bows. So you know they have me.

Happy Holidays!


(1) An 8-core IBM Power 780 (2 chips, 32 threads) with IBM DB2 9.5 is the best 8-core system (1,200,011 tpmC, $.69/tpmC, configuration available 10/13/10) vs. Oracle Database 11g Release 2 Standard Edition One and Oracle Linux on Cisco UCS c250 M2 Extended-Memory Server, 1,053,100 tpmC, $0.58/tpmC, available 12/7/2011.
Source: Results current as of 12/16/11.
TPC-C ,TPC-H, and TPC-E are trademarks of the Transaction Performance Processing Council (TPPC).

The postings on this site solely reflect the personal views of the author and do not necessarily represent the views, positions, strategies or opinions of IBM or IBM management.

technorati tags: , , , , ,,,,,,,,,,,,,


Written by benchmarkingblog

December 16, 2011 at 10:37 am

Posted in Cisco, Oracle, POWER7, TPC-C

Tagged with , , , ,

2 Responses

Subscribe to comments with RSS.

  1. That is not really the goal of this test. It was to showcase low-cost and high performance solutions.
    And your comparison is a bit off due to this difference. IBM would need to configure x3650 with DB2 Express Edition to provide a similar platform. Your reference to the p780 is not relevant in this case.
    Interesting to note that IBM does appear to have recent TPC benchmark listed that has solution price less than 1 million dollars.

    If you are a recent startup company looking to buy infrastructure would you spend that money on specialized IBM gear (p Series servers, AIX and DB2 EE) or would you use something like a 2 Socket Intel server running the most cost effective DB you can find (Oracle Standard Edition 1, MongoDB, My SQL etc)


    December 16, 2011 at 5:19 pm

  2. Elisabeth,

    You mentioned the Power 780 TPC-C benchmark (, so I have a few questions…

    Is there an explanation for why this benchmark used only a 2 socket / 8 processor core configuration, when the Power 780 could have scaled to 8 sockets and 64 processor cores? Was it to leverage the system bandwidth of a high-end server with an entry-level CPU configuration?

    It would be interesting to see how the power 780 would have scaled to the maximum configuration at 4.14GHz of 8 sockets and 32 processor cores? Or alternatively, 8 sockets and 64 processor cores at 3.86 GHz, all be it with 50% less L3 cache per core with 8 processor cores per socket 3.86GHz model?

    I’m presuming IBM wouldn’t want too much interpretation to be taken from the TPC-C clustered benchmark of 3 x Power 780 (24 sockets / 192 processor cores) @3.86GHz, since the performance per core for this benchmark was 65% lower than the 2 socket / 8 processor cores benchmark at 4.14GHz. What’s the scalability issue here? The Power 780’s system bandwidth? Cache in the 8 core per socket configuration? AIX? DB2?

    Either way, is it correct both the 8 socket / 32 core 4.14GHz and 8 socket / 64 core 3.86GHz configurations occupy the same 16u of rack space? If so, presumably this is because the 4.14 GHz config. is physically 64 cores with only 32 cores active? This doesn’t seem very efficient for ISV software licensing where the licensing rules are per physical core. In such a scenario, the 4.14GHz configuration would offer marginally higher performance per core, yet with almost 2x higher software licensing costs per core?

    Also, how would the new Power 780 ‘C’ models be positioned here? I see the ‘C’ variant can scale to 16 sockets and 96 processor cores and the IBM rPerfs are as follows:

    GHz / Sockets / Processor Cores / rPerfs

    4.14 / 8 / 32 / 425.50
    3.92 / 8 / 64 / 692.5
    3.44 / 16 / 96 / 886.6

    Is the 3.44GHz rPerf of 886.6 a typo? Do customers really only get a 28% performance increase from 50% more processor cores when the new 16 socket model is compared to the 3.92GHz config.? Once again, this isn’t a very efficient approach to investing in software licensed per processor core.

    Furthermore, is the new Power 780 3.44GHz model a signal of intent from IBM to Power customers that future scalability and power and space efficiencies will only be possible at lower CPU performance and / or reduced clock speeds?


    January 4, 2012 at 9:59 am

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: