Hello, I am facing a strange problem with Java RMI and Xen. I will describe the problem as well as the experimental results. Setup: I have taken 2 machines, machine1 and machine2. Machine1 is the client and Machine2 is the server. The RMI registry is available on Machine2. I have a RMI application (a kind of service) that is running on Machine2 on port 4001. An init script starts the service at boot time on port 4001. Scenario1: Machine2 is booted with xen as the kernel The client on Machine1 tries to look up the object that it needs to contact in the registry on Machine2. The object to be available in the RMI registry on Machine2 takes around 2 minutes. Scenario2: Machine2 is booted with Linux kernel (Xen is not involved at all) The client on Machine1 tries to look up the object that it needs to contact in the registry on Machine2. The object to be available in the RMI registry on Machine2 takes around 21 seconds. In scenario1 I am not starting any virtual machines. The service is available in Domain0 (Machine2). Nothing is changed except the kernel to boot. But, the results look strange. The Xen and Java RMI stuff looks very strange. The problem here is, on a Xen machine the Java object to be available in the RMI registry takes very long time. According to our experiments it shows that it is approximately 6 times slow. Is it normal with RMI-Xen? Can any body give pointers to this problem? Thanks, Sai Sai S. Dharanikota Intel GmbH Software & Solutions Group Hermuelheimer Strasse 8a 50321 Bruehl Germany Tel : +49 2232 209053 _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> In scenario1 I am not starting any virtual machines. The > service is available in Domain0 (Machine2). Nothing is > changed except the kernel to boot. But, the results look > strange. The Xen and Java RMI stuff looks very strange. The > problem here is, on a Xen machine the Java object to be > available in the RMI registry takes very long time. According > to our experiments it shows that it is approximately 6 times > slow. Is it normal with RMI-Xen?Sounds like you need to do some more basic network performance measurements to figure out if there''s a general problem with networking under Xen on your box. E.g. ttcp or netperf. Cheers, Ian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> I am facing a strange problem with Java RMI and Xen. I will describe the > problem as well as the experimental results.Yup, it''s definitely a strange looking problem!> Scenario1: Machine2 is booted with xen as the kernelHow much memory have you configured for dom0? If it''s less than the full allocation (e.g. if you are reserving space for other VMs) then the JVM / RMI Registry could be thrashing the swap device? If it''s not that then, as Ian said, some other network performance numbers (e.g. ttcp, or just ftp a large file across and calculate the bandwidth) would be interesting. What version of Xen are you running on? -stable, -testing or -unstable. If you''re running -unstable, what revision are you on? Cheers, Mark> Scenario2: Machine2 is booted with Linux kernel (Xen is not involved at > all) > > The client on Machine1 tries to look up the object that it needs to > contact in the registry on Machine2. The object to be available in the > RMI registry on Machine2 takes around 21 seconds. > > > > In scenario1 I am not starting any virtual machines. The service is > available in Domain0 (Machine2). Nothing is changed except the kernel to > boot. But, the results look strange. The Xen and Java RMI stuff looks > very strange. The problem here is, on a Xen machine the Java object to > be available in the RMI registry takes very long time. According to our > experiments it shows that it is approximately 6 times slow. Is it normal > with RMI-Xen? > > Can any body give pointers to this problem? > > > > Thanks, > > Sai > > > > > > Sai S. Dharanikota > > Intel GmbH > > Software & Solutions Group > > Hermuelheimer Strasse 8a > > 50321 Bruehl > > Germany > > Tel : +49 2232 209053_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Hello, -----Original Message----- From: Mark Williamson [mailto:mark.williamson@cl.cam.ac.uk] Sent: Tuesday, April 26, 2005 9:53 PM To: xen-devel@lists.xensource.com Cc: Dharanikota, Sai Subject: Re: [Xen-devel] JavaRMI and Xen problem> I am facing a strange problem with Java RMI and Xen. I will describethe> problem as well as the experimental results.Yup, it''s definitely a strange looking problem!> Scenario1: Machine2 is booted with xen as the kernelHow much memory have you configured for dom0? If it''s less than the full allocation (e.g. if you are reserving space for other VMs) then the JVM / RMI Registry could be thrashing the swap device? I have tested with 128M, 170M, 200M. The results of all the tests look the same. If it''s not that then, as Ian said, some other network performance numbers (e.g. ttcp, or just ftp a large file across and calculate the bandwidth) would be interesting. I have calculated the throughput with the tool netperf. Throughput (10^6 bits/sec) With xen as kernel: 94.13 Without xen as kernel: 94.13 What version of Xen are you running on? -stable, -testing or -unstable. If you''re running -unstable, what revision are you on? I am using Xen 2.0.5 stable binary release ------------------------------------------------------------------- According to my observation the network connection between the two systems are fine. There is no problem with the network. I have done another test which would be very interesting. In this scenario there is only one machine. The client and the server applications run on one machine. Results: Memory 128M 170M 200M With xen as kernel 99.67 sec 112 sec 108 sec Without xen as kernel 13.5 sec 13.5 sec 13.5 sec>From the above results it is pretty clear that the problem is the sameirrespective of how the client and the server are connected. Can any body give pointers to my problem? Do I need to do any additional tests? Thanks, Sai _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> From the above results it is pretty clear that the problem is the same > irrespective of how the client and the server are connected. Can any > body give pointers to my problem? Do I need to do any additional tests?OK, scratch the network performance as a potential problem then. Are you running this in dom0 or in a domU? If it''s a domU, how are you backing its storage? How do other benchmarks perform in the domain - e.g. Linux kernel compile or your favourite system benchmark? Is everything slow, or just the RMI stuff? Do you get the same results with the -testing tree? Cheers, Mark _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
-----Original Message----- From: Mark Williamson [mailto:mark.williamson@cl.cam.ac.uk] Sent: Wednesday, April 27, 2005 3:15 PM To: Dharanikota, Sai Cc: xen-devel@lists.xensource.com; Ian Pratt Subject: Re: [Xen-devel] JavaRMI and Xen problem> From the above results it is pretty clear that the problem is the same> irrespective of how the client and the server are connected. Can any> body give pointers to my problem? Do I need to do any additionaltests? OK, scratch the network performance as a potential problem then. Are you running this in dom0 or in a domU? If it''s a domU, how are you backing its storage? I am not creating any virtual machines. I am using only dom0. How do other benchmarks perform in the domain - e.g. Linux kernel compile or your favourite system benchmark? Is everything slow, or just the RMI stuff? I haven''t tested any other benchmarks. But I will give a try. Do you get the same results with the -testing tree? I am not using the testing tree. Cheers, Mark Cheers, Sai _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> I am not creating any virtual machines. I am using only dom0.OK.> How do other benchmarks perform in the domain - e.g. Linux kernel > compile or > > your favourite system benchmark? Is everything slow, or just the RMI > stuff? > > > > I haven''t tested any other benchmarks. But I will give a try.Cheers.> Do you get the same results with the -testing tree? > > > > I am not using the testing tree.I mentioned it because the testing tree contains some block performance fixes for domUs when using a Linux backend. If you''re just using dom0 I don''t know of anything else in the testing tree that would be relevant to this problem. Cheers, Mark _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Hello, I have made a test with another java application: a simple test program. It works fine. I don''t see any problem. I think the problem is some thing specific to RMI-Xen and also concerned to how we start the RMI application. I see the problem only when the client and the server applications are started by the init script. When I try to start the client and the server applications manually, I don''t see any problem. Few months back, I have tested "povray" benchmark. I remember that I didn''t face any problem when I try to run povray benchmark application in domain0. Can you please suggest how should I proceed? Do I need to do any additional tests? Thanks, Sai -----Original Message----- From: Mark Williamson [mailto:mark.williamson@cl.cam.ac.uk] Sent: Wednesday, April 27, 2005 3:41 PM To: Dharanikota, Sai Cc: xen-devel@lists.xensource.com; Ian Pratt Subject: Re: [Xen-devel] JavaRMI and Xen problem> I am not creating any virtual machines. I am using only dom0.OK.> How do other benchmarks perform in the domain - e.g. Linux kernel > compile or > > your favourite system benchmark? Is everything slow, or just the RMI > stuff? > > > > I haven''t tested any other benchmarks. But I will give a try.Cheers.> Do you get the same results with the -testing tree? > > > > I am not using the testing tree.I mentioned it because the testing tree contains some block performance fixes for domUs when using a Linux backend. If you''re just using dom0 I don''t know of anything else in the testing tree that would be relevant to this problem. Cheers, Mark _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel