On Tue, Apr 21, 2020 at 05:24:01PM +0100, Anton Ivanov wrote: > > On 21/04/2020 17:04, William Tu wrote: > >On Tue, Apr 21, 2020 at 09:23:57AM +0100, anton.iva...@cambridgegreys.com > >wrote: > >>From: Anton Ivanov <anton.iva...@cambridgegreys.com> > >> > >>We do not check metadata - mtime, atime, anywhere, so we > >>do not need to update it every time we sync the log. > >>if the system supports it, the log update should be > >>data only > >> > >>Signed-off-by: Anton Ivanov <anton.iva...@cambridgegreys.com> > >LGTM, > >But how do you know we do not check mtime or atime of the ovsdb log file? > > By searching through the code :) > > stat and stat64 are fairly easy to grep for and so is their use. They are > mostly confined to ssl-stream.c > > >If there isn't a lot of performance overhead updating the metadata, > >why not keep it as it is now? > > The performance overhead on a spinning rust disk is massive - it is several > times. In fact, you can hear it. My source trees are on a SAS 7.2K RPM array > and with the current upstream the testsuite sounds like a hectic Call of Duty > bout. > > With this, the sound becomes much "smoother" - you can hear that the disks > are no longer searching like crazy. > > On a non-rotational - not so much, but still in the 10s of percent. The > kernel has to issue at least two sets of requests instead of one and there is > quite a bit of locking and barriers on metadata updates. > > A. Applied to master, thanks.
_______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev