RE: Performance (was Re: [JBoss-user] JBoss verses Orion)
Tom coleman was working with a guy from SUN directly. We can ping them ... how is ECPerf catching on in the market place I am not hearing anything about it... in fact is it still alive? marc |-Original Message- |From: [EMAIL PROTECTED] |[mailto:[EMAIL PROTECTED]]On Behalf Of James Cook |Sent: Monday, April 02, 2001 1:33 PM |To: [EMAIL PROTECTED] |Subject: Re: Performance (was Re: [JBoss-user] JBoss verses Orion) | | |Marc, | |Have you guys published the appropriate deployment descriptors to use with |ECPerf yet? So we can try it out. | |jim | |- Original Message - |From: "marc fleury" [EMAIL PROTECTED] |To: [EMAIL PROTECTED] |Sent: Monday, April 02, 2001 12:38 PM |Subject: RE: Performance (was Re: [JBoss-user] JBoss verses Orion) | | |One word... | |ECPerf, | |the rest is BS and we can't go down and look at every little "home brew" |test going on. | |marc | | | ||-Original Message- ||From: [EMAIL PROTECTED] ||[mailto:[EMAIL PROTECTED]]On Behalf Of danch ||Sent: Monday, April 02, 2001 10:56 AM ||To: [EMAIL PROTECTED] ||Subject: Re: Performance (was Re: [JBoss-user] JBoss verses Orion) || || ||Rickard berg wrote: || || danch wrote: || || I've spent the day running some test similiar to Paul's. || || My environment is Linux on a PIII 500, running Sun's 1.3 and ||1.3.1beta JDKs. || || || My environment is Athlon 650MHz, Sun 1.3, HotSpot Client (i.e. standard || install). || || || In the case of a standalone client looping and repeatedly calling the || same method on a stateless session bean, I found that Orion averaged || about 3ms. per invocation where JBoss was averaging 6-10 ms/invocation. || || || With the HelloWorld test in the "jbosstest" module I get 3ms/call, || consistently. ||What concerns me isn't the absolute number, but the differential. ||Comparing how fast your machine runs to how fast mine runs doesn't tell ||us much, really. How fast does Orion run HelloWorld on your machine? || || || || I also made a servlet which performs an equivalent test in-VM. This was || even more disturbing, with Orion taking 10-50 ms to perform 5000 || invocations and JBoss/Tomcat taking 2250 or thereabouts. This is such a || huge difference that I have to believe that optimized invocations are || broken in my configuration. ||Well, I was wrong there - it does correctly optimize calls. || || || || Running the Tomcat test EAR, which does in-VM calls to a HelloWorld type || bean I get 0.08ms per invocation, i.e. 400ms for 5000 calls. ||Which would still be over 8 times longer than orion. || || || I don't know what's responsible for our differences, but the ones you || quote point to either bad environment or bad configuration (since JBoss || *can* do much better). ||Bad environment would handicap Orion as well (although it may be that it ||would be handicapped differently. || ||-danch || || ||___ ||JBoss-user mailing list ||[EMAIL PROTECTED] ||http://lists.sourceforge.net/lists/listinfo/jboss-user | | |___ |JBoss-user mailing list |[EMAIL PROTECTED] |http://lists.sourceforge.net/lists/listinfo/jboss-user | | | |___ |JBoss-user mailing list |[EMAIL PROTECTED] |http://lists.sourceforge.net/lists/listinfo/jboss-user ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user
Re: Performance (was Re: [JBoss-user] JBoss verses Orion)
It is still considered "early release". http://java.sun.com/j2ee/ecperf/ The JSR's public review is closed. I think we will see a lot more of it when it is finalized, but now is the time to start getting instructions together for deployment. This will entail the same items that you have provided for the PetStore demo. Hopefully just some deployment descriptors, and a program/script to creat/populate database tables. It is my belief that there is a great disparity in performance among the many ejb containers. I think ECPerf will be a huge success if it is easily deployed and if it works! ;-) jim - Original Message - From: "marc fleury" [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Monday, April 02, 2001 2:08 PM Subject: RE: Performance (was Re: [JBoss-user] JBoss verses Orion) Tom coleman was working with a guy from SUN directly. We can ping them ... how is ECPerf catching on in the market place I am not hearing anything about it... in fact is it still alive? marc |-Original Message- |From: [EMAIL PROTECTED] |[mailto:[EMAIL PROTECTED]]On Behalf Of James Cook |Sent: Monday, April 02, 2001 1:33 PM |To: [EMAIL PROTECTED] |Subject: Re: Performance (was Re: [JBoss-user] JBoss verses Orion) | | |Marc, | |Have you guys published the appropriate deployment descriptors to use with |ECPerf yet? So we can try it out. | |jim | |- Original Message - |From: "marc fleury" [EMAIL PROTECTED] |To: [EMAIL PROTECTED] |Sent: Monday, April 02, 2001 12:38 PM |Subject: RE: Performance (was Re: [JBoss-user] JBoss verses Orion) | | |One word... | |ECPerf, | |the rest is BS and we can't go down and look at every little "home brew" |test going on. | |marc | | | ||-Original Message- ||From: [EMAIL PROTECTED] ||[mailto:[EMAIL PROTECTED]]On Behalf Of danch ||Sent: Monday, April 02, 2001 10:56 AM ||To: [EMAIL PROTECTED] ||Subject: Re: Performance (was Re: [JBoss-user] JBoss verses Orion) || || ||Rickard berg wrote: || || danch wrote: || || I've spent the day running some test similiar to Paul's. || || My environment is Linux on a PIII 500, running Sun's 1.3 and ||1.3.1beta JDKs. || || || My environment is Athlon 650MHz, Sun 1.3, HotSpot Client (i.e. standard || install). || || || In the case of a standalone client looping and repeatedly calling the || same method on a stateless session bean, I found that Orion averaged || about 3ms. per invocation where JBoss was averaging 6-10 ms/invocation. || || || With the HelloWorld test in the "jbosstest" module I get 3ms/call, || consistently. ||What concerns me isn't the absolute number, but the differential. ||Comparing how fast your machine runs to how fast mine runs doesn't tell ||us much, really. How fast does Orion run HelloWorld on your machine? || || || || I also made a servlet which performs an equivalent test in-VM. This was || even more disturbing, with Orion taking 10-50 ms to perform 5000 || invocations and JBoss/Tomcat taking 2250 or thereabouts. This is such a || huge difference that I have to believe that optimized invocations are || broken in my configuration. ||Well, I was wrong there - it does correctly optimize calls. || || || || Running the Tomcat test EAR, which does in-VM calls to a HelloWorld type || bean I get 0.08ms per invocation, i.e. 400ms for 5000 calls. ||Which would still be over 8 times longer than orion. || || || I don't know what's responsible for our differences, but the ones you || quote point to either bad environment or bad configuration (since JBoss || *can* do much better). ||Bad environment would handicap Orion as well (although it may be that it ||would be handicapped differently. || ||-danch || || ||___ ||JBoss-user mailing list ||[EMAIL PROTECTED] ||http://lists.sourceforge.net/lists/listinfo/jboss-user | | |___ |JBoss-user mailing list |[EMAIL PROTECTED] |http://lists.sourceforge.net/lists/listinfo/jboss-user | | | |___ |JBoss-user mailing list |[EMAIL PROTECTED] |http://lists.sourceforge.net/lists/listinfo/jboss-user ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user
RE: Performance (was Re: [JBoss-user] JBoss verses Orion)
One word... ECPerf, the rest is BS and we can't go down and look at every little home brew test going on. marc |-Original Message- |From: [EMAIL PROTECTED] |[mailto:[EMAIL PROTECTED]]On Behalf Of danch |Sent: Monday, April 02, 2001 10:56 AM |To: [EMAIL PROTECTED] |Subject: Re: Performance (was Re: [JBoss-user] JBoss verses Orion) | | |Rickard Öberg wrote: | | danch wrote: | | I've spent the day running some test similiar to Paul's. | | My environment is Linux on a PIII 500, running Sun's 1.3 and |1.3.1beta JDKs. | | | My environment is Athlon 650MHz, Sun 1.3, HotSpot Client (i.e. standard | install). | | | In the case of a standalone client looping and repeatedly calling the | same method on a stateless session bean, I found that Orion averaged | about 3ms. per invocation where JBoss was averaging 6-10 ms/invocation. | | | With the HelloWorld test in the jbosstest module I get 3ms/call, | consistently. |What concerns me isn't the absolute number, but the differential. |Comparing how fast your machine runs to how fast mine runs doesn't tell |us much, really. How fast does Orion run HelloWorld on your machine? | | | | I also made a servlet which performs an equivalent test in-VM. This was | even more disturbing, with Orion taking 10-50 ms to perform 5000 | invocations and JBoss/Tomcat taking 2250 or thereabouts. This is such a | huge difference that I have to believe that optimized invocations are | broken in my configuration. |Well, I was wrong there - it does correctly optimize calls. | | | | Running the Tomcat test EAR, which does in-VM calls to a HelloWorld type | bean I get 0.08ms per invocation, i.e. 400ms for 5000 calls. |Which would still be over 8 times longer than orion. | | | I don't know what's responsible for our differences, but the ones you | quote point to either bad environment or bad configuration (since JBoss | *can* do much better). |Bad environment would handicap Orion as well (although it may be that it |would be handicapped differently. | |-danch | | |___ |JBoss-user mailing list |[EMAIL PROTECTED] |http://lists.sourceforge.net/lists/listinfo/jboss-user ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user
Performance (was Re: [JBoss-user] JBoss verses Orion)
I've spent the day running some test similiar to Paul's. My environment is Linux on a PIII 500, running Sun's 1.3 and 1.3.1beta JDKs. In the case of a standalone client looping and repeatedly calling the same method on a stateless session bean, I found that Orion averaged about 3ms. per invocation where JBoss was averaging 6-10 ms/invocation. There was no logging output from either app. server or the beans. Something interesting about the spread of numbers in JBoss' case is that the low numbers were where it was running off of a local drive, whereas the high numbers where off of an NFS share. What prompted me to make that shift (from NFS in my home to a local /usr/local) was seeing some classloader and unix filesystem activity with very high 'count' values in a hprof output (from 'java -Xrunhprof' that is). At any rate, it seems that something in JBoss might be hitting disk a bit more than i'd expect once it gets into a steady running state: shouldn't classes be loaded at that point? Is it possible that some configuration of classloaders is still searching jar files even when the class is loaded? Paul, had you written tests where the bean actually did work? I'm curious to see if the ~3-6 second differention in invocation times is a constant or not. I also made a servlet which performs an equivalent test in-VM. This was even more disturbing, with Orion taking 10-50 ms to perform 5000 invocations and JBoss/Tomcat taking 2250 or thereabouts. This is such a huge difference that I have to believe that optimized invocations are broken in my configuration. At this point, I'm about at the end of what I can find from the standard runhprof output. I'm trying to get JProbe installed, but that isn't going all that well. I should be able to post my client and bean tommorrow, if anyone is curious. thanks, danch Fink, Paul wrote: Oh sorry I should have said. Jboss 2.1 binary (down loaded March 22) Orion is 1.3.8 performance dependent the JBoss version ran very slow. After playing with cache setting, etc. It seemed like the communications was the main bottle neck. To test this I wrote a simple session bean with one get method that returns a Long. The client looped 10,000 times calling the getter. The Orion version was 6 times faster! Other tests we ran had Orion running 4x faster. It seems that JBoss certainly is performance limited. I'm running on a 900 MHz PIII under Linux with Sun's JDK 1.3 ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user
RE: [JBoss-dev] Performance (was Re: [JBoss-user] JBoss verses Orion)
And who after this will have the guts to argue that we don't need a full time dedicated core that professionally integrates and QAs the code. It was running beautifully not so long ago... I rest my case. Stuff like this is killing us, makes us look like amateurs...the answer to our problem is $$$ And I am working on it, as hard as i can... marc PS: The Apache web server went through the same. |-Original Message- |From: [EMAIL PROTECTED] |[mailto:[EMAIL PROTECTED]]On Behalf Of marc |fleury |Sent: Saturday, March 31, 2001 11:09 PM |To: danch; [EMAIL PROTECTED]; |[EMAIL PROTECTED] |Subject: RE: [JBoss-dev] Performance (was Re: [JBoss-user] JBoss verses |Orion) | | ||I've spent the day running some test similiar to Paul's. || ||My environment is Linux on a PIII 500, running Sun's 1.3 and ||1.3.1beta JDKs. || ||In the case of a standalone client looping and repeatedly calling the ||same method on a stateless session bean, I found that Orion averaged ||about 3ms. per invocation where JBoss was averaging 6-10 ms/invocation. | |It is very broken we have been running the stuff at well below those |numbers. | |In fact we ran around 1-2 ms (depending on stateless). we will see. | |Pf, tired... | |marc | | |___ |Jboss-development mailing list |[EMAIL PROTECTED] |http://lists.sourceforge.net/lists/listinfo/jboss-development ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user
Re: [JBoss-user] JBoss verses Orion
What version of JBoss? If 2.1 (beta) the latest binary or a build from source? If source, what date? Earlier this year there were some performance problems stomped, but I haven't done any testing for performance. On Fri, 30 Mar 2001, Fink, Paul wrote: I ported our Orion application to JBoss. I love JBoss for usability, documentation and support. Unfortunately our application is very performance dependent the JBoss version ran very slow. After playing with cache setting, etc. It seemed like the communications was the main bottle neck. To test this I wrote a simple session bean with one get method that returns a Long. The client looped 10,000 times calling the getter. The Orion version was 6 times faster! Other tests we ran had Orion running 4x faster. It seems that JBoss certainly is performance limited. I'm running on a 900 MHz PIII under Linux with Sun's JDK 1.3 Paul Fink ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user -- Dan Christopherson (danch) nVisia Technical Architect (www.nvisia.com) Opinions expressed are mine and do not neccessarily reflect any position or opinion of nVISIA. --- If you're a capitalist and you have the best goods and they're free, you don't have to proselytize, you just have to wait. -Eben Moglen ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user
RE: [JBoss-user] JBoss verses Orion
Oh sorry I should have said. Jboss 2.1 binary (down loaded March 22) Orion is 1.3.8 -Original Message- From: Dan Christopherson To: 'jBoss' Cc: 'Orion-Interest' Sent: 3/30/01 3:21 PM Subject: Re: [JBoss-user] JBoss verses Orion What version of JBoss? If 2.1 (beta) the latest binary or a build from source? If source, what date? Earlier this year there were some performance problems stomped, but I haven't done any testing for performance. On Fri, 30 Mar 2001, Fink, Paul wrote: I ported our Orion application to JBoss. I love JBoss for usability, documentation and support. Unfortunately our application is very performance dependent the JBoss version ran very slow. After playing with cache setting, etc. It seemed like the communications was the main bottle neck. To test this I wrote a simple session bean with one get method that returns a Long. The client looped 10,000 times calling the getter. The Orion version was 6 times faster! Other tests we ran had Orion running 4x faster. It seems that JBoss certainly is performance limited. I'm running on a 900 MHz PIII under Linux with Sun's JDK 1.3 Paul Fink ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user -- Dan Christopherson (danch) nVisia Technical Architect (www.nvisia.com) Opinions expressed are mine and do not neccessarily reflect any position or opinion of nVISIA. --- If you're a capitalist and you have the best goods and they're free, you don't have to proselytize, you just have to wait. -Eben Moglen ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user
Re: [JBoss-user] JBoss verses Orion
I don't think paul was running his test from a web app environment. I've just written an equivelent session bean and client and JBoss seems to be about 3 times slower than orion right now. I'm going to do a little profiling and see if I can see why. Alexander Jerusalem wrote: I'm not completely sure if this is the reason but there is this "one VM/classloader or multiple VMs/classloaders" distinction. I believe the important point is that Orion is an EJB *and* Servlet container whereas JBoss is only an EJB container. So if you use tomcat for example to run your Servlets all the calls to JBoss based EJBs will we marshalled by value by default whereas in Orion they will be transferred by reference. This makes a huge difference especially if the method you call for performance testing does essetially nothing. There is, however, a way to configure the JBoss/tomcat (or JBoss/Jetty) combination so that this boundary is optimized away. With JBoss comes a sample application that you can call in two ways: either optimized or non optimized and the performance difference is aprox. 4 times or so. I think that this could be the difference you are seeing in your tests. You'll have to ask the JBoss experts to find out about how to configure JBoss so that it uses this optimization. The other thing that I've heard is that tomcat is rather slow in compaison to Jetty. And I know that the Orion JSP/Servlet engine is pretty fast so you should maybe try the JBoss/Jetty combination for faster results. Regards, Alexander Jerusalem At 00:44 31.03.01, Fink, Paul wrote: Oh sorry I should have said. Jboss 2.1 binary (down loaded March 22) Orion is 1.3.8 -Original Message- From: Dan Christopherson To: 'jBoss' Cc: 'Orion-Interest' Sent: 3/30/01 3:21 PM Subject: Re: [JBoss-user] JBoss verses Orion What version of JBoss? If 2.1 (beta) the latest binary or a build from source? If source, what date? Earlier this year there were some performance problems stomped, but I haven't done any testing for performance. On Fri, 30 Mar 2001, Fink, Paul wrote: I ported our Orion application to JBoss. I love JBoss for usability, documentation and support. Unfortunately our application is very performance dependent the JBoss version ran very slow. After playing with cache setting, etc. It seemed like the communications was the main bottle neck. To test this I wrote a simple session bean with one get method that returns a Long. The client looped 10,000 times calling the getter. The Orion version was 6 times faster! Other tests we ran had Orion running 4x faster. It seems that JBoss certainly is performance limited. I'm running on a 900 MHz PIII under Linux with Sun's JDK 1.3 Paul Fink ___ JBoss-user mailing list [EMAIL PROTECTED] http://lists.sourceforge.net/lists/listinfo/jboss-user