Oracle leads multinode SPECJAppServer2002 with new results

Discussions

News: Oracle leads multinode SPECJAppServer2002 with new results

  1. Oracle submitted a new SpecjAppServer2002 benchmark number running Oracle Application Server 10g Standard Edition on Sun hardware, in the multinode category.

    Oracle now leads the multinode category in performance with 2408.73 total operations per second (TOPS) with a fairly average submission on price/performance ($700.08/TOPS).

    You can view all the results at:
    http://www.spec.org/osg/jAppServer2002/results/jAppServer2002.html

    You can view the details on this result at:
    http://www.spec.org/jAppServer2002/results/res2003q4/jAppServer2002-20031014-00017.html

    Oracle recently released 2 other results in the multi-node category, running on HP hardware instead of Sun. Those results had a lower TOPS (863 and 1165) with a much lower price/TOPS ($168 and $150).

    This new submission just pips the HP's report of BEA WebLogic running on HP-UX 11i, which had a slightly lower TOPS (2238) and a higher price/TOPS ($1089).

    Threaded Messages (7)

  2. Every company leads somehow.
          I m sure after a couple of days there will be one more posting saying Web sphere leads, then it will be say JBOSS and so on.
    Why do we really report it when we know the environments, machines, apps used for testing, even the back ends ( databases used ) can affect the final result.
    And to compare two results from tow diff. vendors then is no better than comparing apples and bananas.
  3. Follow the numbers[ Go to top ]

    Hmm,

    14 CPUs for Appservers.
    24 CPUs for the Database.

    2409 TOPS at $700.08/TOPS

    But their best TOPS figure:

    6 CPUS for Appserver
    4 CPUS for the Database

    1165 TOPS at $151/TOPS.

    So if I just double my Windows 2000 investment, I get similar throughput at half the costs. Seems like BEA's JRockit is the real scalability player here not the appserver.

    What's the point of this result except for Larry Ellison propaganda that "I win".

    - Frank Bolander
  4. Following numbers[ Go to top ]

    I think that the reason that they stopped at 6+4 cpu:s is that more cpu:s in that configuration did add more cost than performance. So maybe your simplyfied calculation that 2 6+4 system would perform twice as good doesn't hold. The cost of keeping more machinces running together might be much higher.

    I think that you can be very sure that if they could have gotten better perfomance without getting a higher price/TOPS they would have set upp a server farm that would make the combined power of the top500 look small.

    So, if you have to perform 2409 TOPS you will not get it at a cost of 151$. Then you have to settle for 1165 TOPS.
  5. Following numbers[ Go to top ]

    Frank Bolander

    " Seems like BEA's JRockit is the real scalability player here not the appserver."

    Just to clarify JRockit was not used in this submission. Sun's JVM was used in this submission.

    EJB Container:
    --------------
    JVM Name: Java(TM) 2 Runtime Environment, Standard EditionVersion 1.4.2_01

    JVM Vendor: Sun Microsystems, Inc.
  6. Following numbers[ Go to top ]

    Frank Bolander

    > " Seems like BEA's JRockit is the real scalability player here not the appserver."
    >
    > Just to clarify JRockit was not used in this submission. Sun's JVM was used in this submission.
    >
    > EJB Container:
    > --------------
    > JVM Name: Java(TM) 2 Runtime Environment, Standard EditionVersion 1.4.2_01
    >
    > JVM Vendor: Sun Microsystems, Inc.

    Sorry. Let me clarify. The Windows 2000 benchmark was with JRockit.
  7. Following numbers[ Go to top ]

    I think that the reason that they stopped at 6+4 cpu:s is that more cpu:s in that configuration did add more cost than performance. So maybe your simplyfied calculation that 2 6+4 system would perform twice as good doesn't hold. The cost of keeping more machinces running together might be much higher.

    >

    TOPS does not take into account TOC. TOPS is a "simple calculation" proposition. Also, there is an assumption that multinode configurations scale linearly with added resources to the best extent possible. Hence my extraploation. Otherwise, these numbers are even more useless than they already are.

    > So, if you have to perform 2409 TOPS you will not get it at a cost of 151$. Then you have to settle for 1165 TOPS.

    No, I'm saying that for around $300/TOPS I can get 2300 TOPS. TOPS/$ is a normalized metric on the economic efficiency of the resources I'm throwing at a problem. Since the network topologies are identical in the two cases, we can pretty much cross this factor out of the equation. What's left is Oracle 10g on two different platforms with two different JVMs. Oracle's numbers are saying that the Windows 2000 computing resources with BEA JRockit are more economically efficient than an equivalent config using Linux, Solaris and Sun's JVM for the same problem set. The straight throughput number is worthless other than in claiming victory in pissing contests.

    I wish they would rerun their $700 benchmark with JRockit to give all of us a glimpse at efficiencies between JRockit and Sun JVM in the same sandbox.

    Just for the record, I'm not proposing Microsoft as a viable production solution. I can't stand these benchmarks but management looks at numbers like these when they make decisions.

    - Frank Bolander
  8. Following numbers[ Go to top ]

    TOPS does not take into account TOC


    Sorry TCO(Total Cost of Ownership).