I'm doing a performance test of an nfs server using Filebench. The test is meant to emulate the performance of a cache such as squid hitting an nfs server. The nfs server must be able to serve a large file set - tens of millions of roughly 30KB files - as would be the case in a moderate sized photo sharing consumer internet site.
I have filebench setup and it runs fine for small file sets of say 10K files. As soon as the fileset size grows above about 40K files filebench chokes, giving errors such as "out of memory" and "mutex lock failed". The machine running filebench has 2GB of RAM and no other software running. One error indicated that filebench was trying to allocate over 500GB of RAM. So my question is - is filebench just broken or is there some secret configuration that I'm missing? I've tried setting the thread memsize as high as 200m and as low as 10m to no avail. The need to run tests on large file sets seems fairly basic - small file sets get served out of RAM from the target NFS server, so the file set size must exceed the size of the target 2GB RAM machine. Note that both client and nfs server are Linux Fedora Core 7 boxes. Any ideas? This message posted from opensolaris.org
