FYI. I was inspired to write this email by the Lustre BOF at SC06 where several sites presented their Lustre installations. We are using a Lustre product from HP (called HP SFS) since nearly 2 years on an Itanium system with 120 clients and have just installed an Opteron system with 760 clients. We are pretty happy with the stability and performance of the product (thanks to CFS and HP). Several talks about our experiences can be found at http://www.rz.uni-karlsruhe.de/dienste/lustretalks During the Lustre BOF one question was about available monitoring tools. HP has its own proprietary set of tools which were described in my talk at the HP-CCN in Seattle. However, the general concepts could also be used on other systems. A perl script is used to gather performance data below /proc on clients and servers. This data can be directly displayed in text format (or stored in order to create graphical charts). Once we discover high throughput or metadata rates on the servers we often check which clients are creating this high IO load. The batch system finally shows which users are responsible. This allows us to talk with these users and possibly improve their applications. Regards, Roland -- ---------------------------------------------------------------------------- Roland Laifer Computing Centre (SSCK), University of Karlsruhe, 76128 Karlsruhe, Germany Email: Roland.Laifer@rz.uni-karlsruhe.de, Phone: +49 721 608 4861, Fax: +49 721 32550, Web: www.rz.uni-karlsruhe.de/personen/roland.laifer ----------------------------------------------------------------------------