Benchmarking is difficult to package as it depends a great deal on the hardware used (mem, cpus) and even the network topology to some degree. So we could certainly package high performance linpack (the benchark used for top500.org) but people would still have to twitter around with it almost as much as if they just installed it themselves. At least thats what we have concluded everytime I have heard it brought up...
On 4/29/06, Geoffroy VALLEE <[EMAIL PROTECTED]> wrote:
Hello Neil, Congratulation for the installation of your clusters and few remarks (sorry if i cut some text)... Le Samedi 29 Avril 2006 05:24, Neil Costigan a écrit: > Secondly I wanted to to share my 'success' to the list. > I've now got 2 OSCAR based clusters running using a bunch of old lab > machines > one 20 node for 'work' and one 8 node for testing latest betas etc. > I'm going to try the debian test this week. > Currently can sign off the pre-release 5.0 working fine on Fedora > Core 4. For the debian test, i will be happy to have your feedback, the website is http://oscarondebian.gforge.inria.fr/. The project cannot be used in production but i am definitively interested in users feedback and of course by contributions. Actually i am still working on the integration of OOD into OSCAR trunk but a lot of stuff still have to be done. Therefore if you want to contribute (even for simple tasks), just let me know. :-) > > lastly > there was a question a week or so again about benchmarking > the answer was that it depends on the intended use. fair enough. > I was thinking that there still must be something more generic that > could help if one wanted to know what improvements (or not) if one say > moved from a hub to a switch, or added memory to the compute nodes, > or doubled the number of nodes > my example would be if i joined some of my 8 test nodes into the main > 'production' setup. > what woud i gain 'in theory' for my application. > any ideas ? I think there are two different points: benchmarking and application gains. For application gains, it depends on your application. If the application is designed to scale up to the new size of your cluster, you should have a speed up. At the end this is the old problem of parallelism. :-) For benchmarking this is another question. Benchmarks is based on well-known algorithms which are often design to scale and take a full benefit of cluster resources. But that does not mean your applications will have the same behavior (it depends on your applications again). Anyway, there are no benchmarks (for what i know) shipped with OSCAR. I think about it since a while but i did not have the time to work on that. If i am not clear or if you want more details, feel free to contact me. Thanks, -- Geoff ------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmdlnk&kid0709&bid&3057&dat1642 _______________________________________________ Oscar-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/oscar-devel
------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid0709&bid&3057&dat1642 _______________________________________________ Oscar-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/oscar-devel
