Hi, For rand small write, libaio or posix direct aio would be very nice for the IOPS, I think this kind of journal should be the same. For data read and write, libaio will perform very nice for rand small read and write. So If user create a rbd image for a KVM image, IOPS will be much better if OSD handle IO with libaio. Bandwidth will be a little better too. Thx Simon 2011/8/20 Sage Weil <sage@xxxxxxxxxxxx>: > On Sat, 20 Aug 2011, huang jun wrote: >> hi,all >> Recently,we have a test to compare async i/o and sync i/o in OSD. >> The osd jounral was written in sync + direct io mode now, that heavily >> infuluenced the write performance. >> we use linux aio API. test file size is 40MB,and disk isn't busy. >> here is the test result: >> type ext3 >> sync write+direct io 4.441s >> >> async write +direct io 0.462s >> >> we are glad to see that, so we want to affirm that : >> 1) can we use async+direct io to achieve better write performance if >> we used osd journal? > > Do you mean changing the writes in os/FileJournal.cc to be async? As long > as the journal does not acknowledge a write until it is stable on disk, > that is fine. Async IO would be a nice improvement here; I think > currently it's a single sync O_DIRECT io in flight at a time. > > The test is to write a 40MB file via which interface? Via a kernel client > mount? > > Thanks- > sage > > >> 2) does this bring some problems? >> >> thank you in advance! >> -- >> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in >> the body of a message to majordomo@xxxxxxxxxxxxxxx >> More majordomo info at http://vger.kernel.org/majordomo-info.html >> >> > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html