New BEA WebLogic JRockit JVM Benchmarks on IA32 and IA64

Discussions

News: New BEA WebLogic JRockit JVM Benchmarks on IA32 and IA64

  1. New SPECjbb benchmark numbers have been published for BEA WebLogic JRockit, on different systems.

    The results:

    Bull NovaScale 4040 (4-way IPF) - 64-bit JRockit 1.4.1 JVM on Windows 2003 - 106,451 ops/s
    * fastest 64-bit JVM on any 4-way system

    Dell PowerEdge 3250 (2-way IPF) - 64-bit JRockit 1.4.1 JVM on Windows 2003 - 54,617 ops/s
    * fastest 64-bit JVM on any 2-way system

    Dell PowerEdge 6650 (4-way Xeon) - 32-bit JRockit 1.4.1 JVM on Windows 2000 - 94,405 ops/s
    * fastest JVM on Windows (IA32) on a 4-way system

    SPECjbb2000 Benchmark Highlights:

    - Emulates a 3-tier system, the most common type of server-side Java application today.

    - Business logic and object manipulation, the work of the middle tier, predominate.

    - Clients are replaced by driver threads, database storage by binary trees of objects.

    - Increasing amounts of workload are applied, providing a graphical view of scalability.

    Go to:

    BEA WebLogic JRockit results

    SPECjbb2000 Home Page

    BEA seems to have put out quite a push recently on the benchmark front
  2. JRockit Rocks[ Go to top ]

    I have a program in Java which is just a stand alone program. It loops through an XML based data model (data is true objects, not database rows trying to look like objects) to build a mathmatical optimization matrix. I have been looking for which JVM is the flat out fastest at just straight executing a single threaded Java Program, and JRockit seems to be it.

    On my platform, (pentium 2.8 Ghz with 800 Mhz bus, 2 GB memory, Windows 2000 server) I had a reference case that took 4 minutes 15 seconds (100% CPU) with Sun 1.3.1_08. With JRockit running the Sun compiled code it ran in 3 minutes 15 seconds. I also tried the IBM 1.3.0 and while I had heard it too was fast, it ran in a disappointing 6 minutes 23 seconds. Of interest was that with the same code based compiled by the IBM SDK, the JRockit run took only 2 minutes 47 seconds. I'm still trying to figure that one out.

    I know this is just my particular case, but I'm curious what type of results other people have had for just plain algorithmic speed of various JVMs. Also besides the 3 sources I mentioned, is there another Windows JVM available that I'm not aware of.

    Daniel
  3. JRockit 8 is JDK 1.4 based JDK. Try other 1.4 JDKs.
  4. Actually I did try Sun 1.4.1_03 and it performed about the same as 1.3.1 (within a second or two) Actually, it was slightly better (about 20 seconds) when I used the 1.4.1 compiler to generate 1.3 targeted code and ran that under 1.4.1).

    The Sun 1.4.2 have a JVM stack dump that I figure to be a JVM bug since so many other JVMs can execute the program.

    The only IBM 1.4.1 I could find is for Linux. If they have a Windows release I'd love a URL for it.
  5. JRockit Rocks[ Go to top ]

    I have read JRockit whitepaper and the MxN thin thread model may be the best thing thats happened to Java on Linux. I am interested in finding out about configuration -- on 1 gb of ram on a SINGLE CPU(intel) Linux box, how many java threads can my appplication spawn ? Or how many Tomcat request processors can be concurrently executing?

    Also, the benchmarks did not inlcude 1 cpu m/c. Why ? On a SINGLE CPU m/c running Linux , would the JRockit System would revert to Mx1 thin threads ? On a SINGLE CPU Linux box, will I get to see the any improvements in concurrent processing (number of threads the application can spawn) over current Linux JVMs?


    Regards
    -manish
  6. * fastest 64-bit JVM on any 4-way system

    >
    > * fastest JVM on Windows (IA32) on a 4-way system

    If I get it right, these were the only aubmitted results of the benchmark running on a 4 CPU machine.

    Thus they are not only the best but also the worst (submitted) results.

    There are lies, bad lies and benchmarks.
  7. Thus they are not only the best but also the worst (submitted) results.


    OK. I should have read the headline that says "Third Quarter 2003 SPEC JBB2000 Results".
  8. BEA Benchmarking Activity[ Go to top ]

    People have a love/hate relationship with benchmarks, ourselves included. While they are not perfect, they do provide more value than the alternative- no data, or vendor run benchmarks that mysteriously always show their products winning.

    BEA is typically involved in three benchmarks;

    SPECjAppServer for WebLogic Server
    http://www.spec.org/jAppServer2002/results/jAppServer2002.html

    SPECjbb for JRockit
    http://www.spec.org/jbb2000/results/jbb2000.html

    TPC-C for Tuxedo
    http://www.tpc.org/tpcc/results/tpcc_perf_results.asp

    The upside is that these benchmark committees bend over backwards to try to create a fair environment to test the performance and price/performance of these products. It is really unbelievable how many rules, processes and debates go into these forums. The downside is that vendors sometimes find ways to try to manipulate the output through special pricing, features, or other tricks. The good news is that the committees pick up on this and closes the loopholes in the rules. SPEC has come a long way since the old ECperf days.

    Fire away if you have any questions about JRockit, the benchmark processes or these results.

    Eric
    BEA Systems
  9. Good news for everyone buying into BEA's hype. Their support matrix
    does not list Windows 2003, yet these benchmarks were submitted under Windows (server) 2003. While JRocket does in fact list Server 2003 as support for ONLY 64-bit versions, the APP server, client, and all other suites from BEA are not supported on 2003.

    Congratulations BEA for showing us results on an unsupported platform.
    And before you reply with "oh we just haven't updated the web site yet",
    I already called your helpdesk and guess what they told me with regards
    to 2003 support? I'll let you learn that lesson the hard way.

    What a class act.
  10. SPEC has very specific rules about General Availability and supported configurations. The rule is that the entire stack has to be GA for production use within 3 months of the publication of the benchmark. This configuraiton will easily meet that requirement.

    Eric
    BEA Systems
  11. Windows 2003 _is_ supported[ Go to top ]

    Good news for everyone buying into BEA's hype. Their support matrix

    > does not list Windows 2003, yet these benchmarks were submitted under Windows (server) 2003. While JRocket does in fact list Server 2003 as support for ONLY 64-bit versions, the APP server, client, and all other suites from BEA are not supported on 2003.

    SPECjbb2000 is a JVM-only benchmark that does not involve the use of an application server. JRockit _is_ already supported as a GA product on Windows 2003 (as you said correctly) for 64-bit IPF systems.

    In addition, independent of this benchmark, WebLogic Server 8.1 SP1 is also available as a fully supported product on Windows 2003 (for 64-bit IPF systems). It can be downloaded today from http://commerce.bea.com/showproduct.jsp?family=WLS&major=8.1&minor=1