James G. Sack (jim) wrote:
Short version: ext4 is awesome. zfs has absurdly fast metadata operations but falls apart on sequential transfer. xfs has great sequential transfer but really bad metadata ops, like 3 minutes to tar up the kernel.
I wonder how Lustre would compare to all of these. I always notice that performance seems to be a tradeoff between metadata and sequential transfer speeds. Perhaps the problem is that these two things are totally different kinds of data that should be stored separately. Lustre stores the data in some sort of filesystem optimized for things like sequential data transfer and the metadata in a database (I don't recall if it is something like mysql or postgres or a berkeleydb) which is optimized for these small data accesses.
-- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
