On Tue, Feb 25, 2014 at 12:05:58PM -0800, Christoph Hellwig wrote: > So what's protecting the cache hits statistics now? Nothing. And I don't care because the statistic is meaningless from the point of view of tuning xfs_repair performance. > Longer term it might make sense to just port the XFS buffercache from > the kernel to libxfs.. *nod* We badly need to separate the buffer cache into per-AG caches because most of the per-thread operations are isolated to a single AG and so a global hash simply doesn't scale from a lock contention or a table sizing point of view.... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs