On Wed 08-01-20 12:42:59, Theodore Y. Ts'o wrote: > On Wed, Jan 08, 2020 at 04:15:13PM +0530, Ritesh Harjani wrote: > > > Yes, that's a good point. And I'm not opposed to that if it makes the life > > > simpler. But I'd like to see some performance numbers showing how much is > > > writeback using unwritten extents slower so that we don't introduce too big > > > regression with this... > > > > > > > Yes, let me try to get some performance numbers with dioread_nolock as > > the default option for buffered write on my setup. > > I started running some performance runs last night, and the Thanks for the numbers! What is the difference between 'default-1' and 'default-2' configurations (and similarly between dioread_nolock-1 and -2 configurations)? > interesting thing that I found was that fs_mark actually *improved* > with dioread_nolock (with fsync enabled). That may be an example of > where fixing the commit latency caused by writeback can actually show > up in a measurable way with benchmarks. Yeah, that could be. > Dbench was slightly impacted; I didn't see any real differences with > compilebench or postmark. Interestingly, dbench is also fsync(2)-bound workload (because the working set is too small for anything else to actually reach the disk in contemporary systems). But file sizes with dbench are smaller (under 100k) than with fs-mark (1MB) so probably that's what makes the difference. > dioread_nolock did improve fio with > sequential reads; which is interesting, since I would have expected > with the inode_lock improvements, there shouldn't have been any > difference. So that may be a bit of wierdness that we should try to > understand. Yes, this is indeed strange. I don't see anything the DIO read path where dioread_nolock would actually still matter. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
