From:
<jmf...@cn...> - 2005-09-07 17:48:07
|
Hi Adam, Adam Retter wrote: > Have you done much performance testing with eXist? e.g. > Hardware/Software combinations and JVM's? Well, I have tested eXist in quite different platforms since one year=20 and a half, and this is the historical order: * Sun Ultra60 (only one 360MHz processor) with 512MB and a 9GB SCSI disk. * Powerbook G4 500MHz (first generation) with 512MB and 20GB slow IDE dis= k. * Athlon 1.4GHz with 1GB and 500GB SCSI array (with IDE disks) in RAID-3. * Acer Laptop with Intel Centrino 1.5GHz, 512MB and 80GB fast IDE disk. * Athlon 64 3200+ with 2GB and 120GB SATA disk. IBM JVM implementation is veeery fast in their processors (G4, POWER4,=20 G5, POWER5). Sun JVM implementation eats *lots* of memory in Solaris=20 Sparc (twice or three times the used one in Intel platforms). Disk=20 performance and speed is crucial when you are using Sun JVM. I think=20 eXist performance can be improved if the transaction log file is located=20 in a partition in a different disk, because it is being synced each time=20 a transaction starts, or finished due rollback or commit. Also, if you=20 have well defined queries, it is worth disabling default full text=20 indexes over the whole document and creating specific ones (either=20 fulltext or not) for the elements/attributes/text you are querying. And from JVMs point of view, what I have learnt is: * In general, if you can choose between 1.4.2 and 5.0 releases, is=20 better the first one (talking about eXist). Perhaps the last one will=20 improve in the future, but now it tends to eat much more memory and is a=20 bit slower than 1.4.2. * jrockit from BEA is not recommended, because version 1.4.2 is not able=20 to handle large files (bigger than 2GB), and although 5.0 is able to, it=20 has some bugs. As I told in a previous mail, it had a segmentation fault=20 at the beginning of my tests (it bursted!). * Sun JVM is faster in I/O but IBM one wastes less memory. > Also are you using Tomcat as a container and if so which version? >=20 No, I'm not using it so. I'm running it in "server" mode. Best Regards, Jos=C3=A9 Mar=C3=ADa --=20 Jos=C3=A9 Mar=C3=ADa Fern=C3=A1ndez Gonz=C3=A1lez e-mail: jmfernandez@cn= b.uam.es Tlfn: (+34) 91 585 54 50 Fax: (+34) 91 585 45 06 Grupo de Dise=C3=B1o de Proteinas Protein Design Group Centro Nacional de Biotecnolog=C3=ADa National Center of Biotechnology C.P.: 28049 Zip Code: 28049 C/. Darwin n=C2=BA 3 (Campus Cantoblanco, U. Aut=C3=B3noma), Madrid (Spai= n) |