On Wed, Aug 21, 2019 at 09:04:57AM +0800, Joseph Qi wrote: > On 19/8/21 00:08, Theodore Y. Ts'o wrote: > > On Tue, Aug 20, 2019 at 11:00:39AM +0800, Joseph Qi wrote: > >> > >> I've tested parallel dio reads with dioread_nolock, it doesn't have > >> significant performance improvement and still poor compared with reverting > >> parallel dio reads. IMO, this is because with parallel dio reads, it take > >> inode shared lock at the very beginning in ext4_direct_IO_read(). > > > > Why is that a problem? It's a shared lock, so parallel threads should > > be able to issue reads without getting serialized? > > > The above just tells the result that even mounting with dioread_nolock, > parallel dio reads still has poor performance than before (w/o parallel > dio reads). Right, but you were asserting that performance hit was *because* of the shared lock. I'm asking what leading you to have that opinion. The fact that parallel dioread reads doesn't necessarily say that it was because of that particular shared lock. It could be due to any number of other things. Have you looked at /proc/lock_stat (enabeld via CONFIG_LOCK_STAT) to see where the locking bottlenecks might be? - Ted
