6 Replies Latest reply on Sep 13, 2004 5:26 AM by david.wade

    Low JBoss performance on Linux

      Hi,

      when I use Jboss with Linux I noticed a slow performance in comparison to using Jboss with Windows.
      For example starting a Jboss on Windows tooks about 10 sec., same on Linux about 22 sec.
      In addition this test used
      - a good equipped Hardware for the Linux server with 2,8Gz xeon, very fast SCSI discs, fast RAM etc.
      - a usual notebook with 1,5Gz and a slow disk.
      There is almoust the same performace difference for deploying beans or doing some transactions. It seems JBoss always runs much faster on windows, also when using a poor hardware.

      Where are the reasons?

      - JDK implementation on linux?
      - OS? (impl./config)

      Any ideas/solutions?

      Thank you

        • 1. Re: Low JBoss performance on Linux
          genman


          Tip: Use Redhat AS or ES. Or Fedora/Linux 2.6. I also recommend JDK 1.4.2 from Blackdown.org . You may also not be running the exact same JBoss configuration in both places. What does "top" say in Linux?

          • 2. Re: Low JBoss performance on Linux

            Thank you.
            I'm currently using Suse Linux 9.1 with a 2.6.x kernel.
            I tested JDK 1.4.2_03 and Blackdown 1.4. This makes almost no difference.
            It is exact the same JBoss on both places.
            'Top' displays no 'idle' time when JBoss gets started. There is more then enough memory free.
            I set the number of open file (ulimit -n) to a big amount, but with no effekt.
            I get the same slow performance on two more linux servers.

            Advice is very welcome. What slows down JBoss on Linux?

            • 3. Re: Low JBoss performance on Linux
              dhartford

              quick reply - my own tests last year also showed jboss better on Win2000 vs linux. And, it was more evident on dual-processor boxes vs single-processor (talk about a suprise!). But that was win2000 vs redhat 9, dual-boot boxes, with Sun jdk 1.4 (can't remember if it was 1.4.1 or 1.4.2).

              Still, I run on redhat for TCO reasons, and that percentage of performance difference didn't justify the increased TCO (not to mention patch/virus downtimes!).

              • 4. Re: Low JBoss performance on Linux

                Using jboss on a windows server is not an option for me, also when it seems to be faster.

                I did some more tests know:
                - using another Linux dist. (Mandrake) does not bring any performance advantage
                - starting up jboss with jdk1.5rc does not make a nameable difference
                - Some other java apps (e.g. a swing client, jbuilder) are executed very fast with the linux VM, faster than at my windows pc. I just noticed jboss to be very slow with the linux jdk's.

                I read some articles about a slower thread implementation done for linxu/jdk.

                Are there some more experiences around here?



                • 5. Re: Low JBoss performance on Linux
                  ebrain13

                  Hello,

                  I have noticed, the same JBoss 3.2.5 takes no time to startup compared to the one installed on Linux ( Mandrake ).

                  Are there any explicit configuration to be done ?

                  Thanks.

                  • 6. Re: Low JBoss performance on Linux
                    david.wade

                    Its true that in general you will see better performance on WinTel than LinTel. However, the differences are not as dramatic as the comparison of JBoss start-up times would suggest.

                    WinTel especially, non-server based, when not running JBoss as a service, will give the JVM every ounce of CPU it has available at the expense of anything else you might try to do at the same time. LinTel (and any UNIX in general) will allow other processes to have an equal share of the CPU if they require it.

                    I would suggest that you not dwell on the startup time, or things like JSP compile time, but compare the performance when everything is up and running.