There's a couple of options in the object server that are related to how object data is cached (or generally *not*) https://github.com/openstack/swift/blob/master/swift/obj/server.py#L921 At scale on dense nodes it becomes challenging to keep all the filesystem metdata in the page cache, so we've tried a few different tricks and tunings over the years to optimize towards using as much RAM as possible on minimizing the number of seeks "wasted" to pickup filesystem directory and extent information instead of object data. BUT on nodes with more RAM and less objects (or fewer *active* objects) it is definitely possible to tune towards keeping more object data in the page cache. Good luck! -Clay On Fri, Oct 6, 2017 at 2:34 PM, Jialin Liu <jalnliu at lbl.gov> wrote: > Hi, > Is there any existing work that leveraging operating system's page cache > for swift? > like many other parallel file systems, lustre, the IO is cached in buffer > and call is returned immediately to user space. > > Best, > Jialin > > __________________________________________________________________________ > OpenStack Development Mailing List (not for usage questions) > Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev > > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20171006/0c37feff/attachment.html>