HP Posts New SPECjAppServer2002 MultiNode Result

Discussions

News: HP Posts New SPECjAppServer2002 MultiNode Result

  1. HP has posted 2 new SPECjAppServer2002 results in the MultiNode category. The configurations run: BEA WebLogic Server 8.1 SP1 on a HP DL 360 Cluster, running on Redhat Linux.

    The difference between the two results seem to be running different injection rates, and slightly different hardware (e.g. the db was changed to have 4 CPUs from 2CPUs)

    Configuration 1:
    The TOPS recorded was 613.23 (compared to 1753 and 2238 on the other WLS 8.1 configs, and 448.12 on IBMs), and the US$/TOPS was 257.75. The $/TOPS is now the lowest, with IBMs 647.52 being the nearest.

    Configuration 2:
    TOPS: 694.88
    US$/TOPS: 262.87

    Read the HP / BEA MultiMode results
    Configuration 1 results

    Configuration 2 results

    Visit all of the SPECjAppServer2002 results
  2. These benchmarks show BEA WebLogic Server, running on half of the processors as WebSphere, achieving 37% more throughput at 60% lower cost per transaction.This price/performance advantage results from the fact that the complete BEA WebLogic/Red Hat/Oracle/HP system, including all hardware, application server and database software licenses and three years of 24x7 support costs 46% less than the WebSphere/Windows/DB2/IBM stack, yet processes substantially
    more transactions.
  3. There is a note in the disclosure that says "Transaction consistency for those beans specifying field groups was proven by manual analysis of the source code". Can someone explains what that refers to?
  4. Field-groups allow you to specify which fields are loaded from the database together, while deferring the loading of others that may not be used. To ensure transaction consistency, all of the fields accessed within a transaction must be included in the field group.

    This statement was included to acknowledge that the application's source code was examined to guarantee that all the fields accessed were included in the field group definition, assuring transaction consistency.
  5. BEA JRockit JVM[ Go to top ]

    It appears that BEA's JRockit JVM was used for these tests.

    My company uses HP-UX. Maybe we'll look at JRockit.

    We've been running both Webpshere 4.x and WebLogic on HP-UX.
  6. BEA JRockit JVM[ Go to top ]

    AFAIK There isnt a HP-UX version of JRockit.

    -Nick
  7. not HP-UX[ Go to top ]

    The OS used was Red Hat Advanced Server 2.1 QU2

    (not HP-UX)
  8. Some of the criticism of past benchmarks has been that the hardware configurations have been so different that you couldn't make useful comparisons of the software stack. This configuration was put together to try to address this issue.

    A complete analysis of this benchmark and how it stacks up to the similar IBM submission can be found at BEA and HP Benchmark Analysis

    Eric
    BEA Systems
  9. Does RH 2.1 AS use the new threading model?
  10. RHAS 2.1 uses LinuxThreads, not NPTL.
  11. Not JRockIt![ Go to top ]

    They used Suns 1.4.1 as stated by the reports:

    > JVM Version used:
    > Java (TM) 2 Runtime Environment, Standard Edition Version 1.4.1
    > JVM arguments used:
    > -Xms1024m -Xmx1024m

    Lars
  12. Sorry, I was wrong[ Go to top ]

    ... they used JRockIt:

    BEA WebLogic JRockit 8.1 SP1 SDK

    :o)

    Lars
  13. Difference in two submissions[ Go to top ]

    The first submission, with 2 CPUs on the database tier, was put together to keep it similar to the IBM configuration. In that scenerio each had 2 database CPUs. The IBM configuraiton added 2x 4CPU app servers (8 CPUs total). The BEA configuration staurated the database with 2x 2CPU app server (4 CPUs total). This is what is so interesting about the comparison.

    The problem was that even though the BEA numbers were seeing 37% more throughput on half the CPUs as WebSphere, with a 60% lower price/performance, the configuraiton was database bound, so there was still more potential left in the WebLogic tier.

    The second submission added two more CPUs to the database tier to show that we could still squeeze more throughput out of the same app server CPUs.

    It is all explained in the white paper linked to above.

    Eric
    BEA Systems
  14. Where the heck is Oracle?[ Go to top ]

    Oracle runs around saying they are the leader in this space, that 9iAS is the fastest, that they have the most customers, bla, bla, bla. Where are their SPEC #s? I'd like to see them put their money where their mouth is. They sure love those bar charts... why do I suspect they can't touch these IBM or BEA numbers???

    They offered my company free app server licnses. Sure, so they can drain more database CPUs??? No doubt! They can't even give the stuff away.

    Harvey
  15. Difference in two submissions[ Go to top ]

    Well, the results look really like they're the number to beat. It's always good to see more results, because with only a few, it's hard to compare them at all. At least with a good number of results, one can start to draw some trend lines.

    Congrats! I'm looking forward to seeing more ;-)

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Easily share live data across a cluster!
  16. BEA JRockit JVM (vs. HP-UX)[ Go to top ]

    It appears that BEA's JRockit JVM was used for these tests.

    My company uses HP-UX. Maybe we'll look at JRockit.

    We've been running both Webpshere 4.x and WebLogic on HP-UX.

    Well, as someone else pointed out, this was *not* an HP-UX system.

    While it appears that Jrockit will be coming out for HP-UX Itanium systems (64-bit only), I encourage you to check out the Spec submissions for HP's JVM on both PA-RISC and Itanium before you leap to the conclusion that Jrockit is faster.

    Especially when you consider that on Itanium the HP JVM can run in 32-bit mode, which is generally faster due to more efficient use of cache and TLB.

    Further, while Jrockit has some very nice performance analysis tools, HP's JVM has similar features (HPjmeter and HPjtune).
  17. I'd like to see results for these servers:

    1) Sun ONE application server
    2) Borland Enterprise Server, AppServer Edition

    I could not find any published results for these app servers.
  18. Hi,

    can anyone point out the differences between both submissions?

    So far, I just found that the DB Server has two CPUs more in the better result.

    Lars