On Sat, 4 Apr 2009, Theodore Tso wrote: > > Using ext3 data=writeback, your "write big (2GB) file and sync" as the > background workload, and fsync-tester, I was able to reduce the > latency down to under 1 second... Hmm. I can certainly see a very noticeable improvement. I could still see 1-2s fsync() delays for my MUA test (and seem to have seen a 4s one once too). But most were all 'quite noticeable stutters', and very seldom do they go into the 'ooh, that's really painful' stage. And yes, anticipatory seems to be quite noticeably better than cfq here. With cfq I got a few two-second delays on 'ftruncate()' too (probably because of your new serialization code?), and the longest fsync() delay was over 7 seconds. That was definitely solidly in the "painful" category. (Jens - my test-case is not the exact same fsycn() test that Ted uses: it's really just me being in my MUA editing an email, and doing that while : ; do time sh -c "dd if=/dev/zero of=bigfile bs=8M count=256 ; sync; rm bigfile"; done in another window. My MUA does these save-files ever minute or so, and does a 'fsync()' after writing that (small) file. If the fsync() takes more than half a second, I can definitely notice. If it takes 1-2 seconds, it's a big stutter, where keyboard auto-repeat when moving around really hurts (ie I don't see how it moves). If it takes 5+ seconds, it feels really bad, and as if the whole email client had just died. Yeah, it's a nasty test. Linus -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
