I read a review of a few parallel file systems the other day (I tried to find the link and couldn’t find it, sorry), but it suggested PVFS, Lustre and GlusterFS. I had used PVFS back in the 1.0-days and thought that it was worth trying again.

I installed 8 vmware instances of Ubuntu 9.10 on a machine that I had lying around and started compiling PVFS. I got the server running on all 8 instances and got the kernel module running on one node, so I had a real, mountable filesystem. I’m not sure how to measure performance because it’s all done in a virtual machine on a 100Mbps network in my home, but I was surprised to find that PVFS was not fault tolerant if/when I took a machine down. What?

I dug through the docs and it turns out that PVFS is only fault tolerant for the server if shared storage is available and it uses heartbeat to do the failover, so the ‘server’ is fault tolerant, but the file system isn’t if we lose a machine or harddrive. Bummer.

Oh well. So much for PVFS. Although it has matured quite a lot since the 1.0 days, it’s still not a ‘redundant’ solution if you need replication and want to design around storage hardware failures.

Tweet