Derek Siu <golf1123@xxxxxxxxx> wrote: > I did some testing on the performance of CacheFS with NFS by copying a > whole bunch of files from the NFS server to a client. Without CacheFS > on, it took around 70 secs. However, with CacheFS on, it took around > 147 seconds!! An entry in 2005 stated that CacheFS suffered from the > fragmentation problem. I wondered if that's the problem....... You aren't using CacheFS, you're using CacheFiles which doesn't suffer from CacheFS's fragmentation problem. CacheFS implemented its own filesystem and used a blockdev as the data store; CacheFiles uses files on some other filesystem that's already mounted. This means that CacheFiles performance is subject to the underlying filesystem. Is it ext3? If so, you might want to check the block size: tune2fs -l /dev/sda5 | grep "Block size" If it's 1KB, you might want to use 4KB instead. Also, if, say, you're NFS connection is running over GigE, then you're almost certainly going to be worse off with a cache. You have to decide from your collection of circumstances whether a cache will be a gain or a loss. David
