Yes, they're both delayed and a guesstimate: the OSDs send periodic information to the monitors about the state of their PGs, which includes amount of data read/written from them. The monitor extrapolates the throughput at each report interval based on the pg updates it received during that time. -Greg On Tue, Sep 15, 2015 at 1:08 PM, Barclay Jameson <almightybeeij@xxxxxxxxx> wrote: > Good point. I have seen some really weird numbers something like 7x my > normal client IO. This happens very rarely though. > > On Tue, Sep 15, 2015 at 2:25 PM, Mark Nelson <mnelson@xxxxxxxxxx> wrote: >> FWIW I wouldn't totally trust these numbers. At one point a while back I >> had ceph reporting 226GB/s for several seconds sustained. While that would >> have been really fantastic, I suspect it probably wasn't the case. ;) >> >> Mark >> >> >> On 09/15/2015 11:25 AM, Barclay Jameson wrote: >>> >>> Unfortunately, it's not longer idle as my CephFS cluster is now in >>> production :) >>> >>> On Tue, Sep 15, 2015 at 11:17 AM, Gregory Farnum <gfarnum@xxxxxxxxxx> >>> wrote: >>>> >>>> On Tue, Sep 15, 2015 at 9:10 AM, Barclay Jameson >>>> <almightybeeij@xxxxxxxxx> wrote: >>>>> >>>>> So, I asked this on the irc as well but I will ask it here as well. >>>>> >>>>> When one does 'ceph -s' it shows client IO. >>>>> >>>>> The question is simple. >>>>> >>>>> Is this total throughput or what the clients would see? >>>>> >>>>> Since it's replication factor of 3 that means for every write 3 are >>>>> actually written. >>>>> >>>>> First lets assume I have only one cephfs client writing data. >>>>> >>>>> If this is total throughput then to get the maximum throughput for >>>>> what a client would see do I need to divide it by 3? >>>>> >>>>> Else, if this is what my client sees then do I need to multiply this >>>>> by 3 to see what my maximum cluster throughput would be? >>>> >>>> >>>> I believe this is client-facing IO. It's pretty simple to check if >>>> you've got an idle cluster; run rados bench and see if they're about >>>> the same or about three times as large. ;) >>>> -Greg >>> >>> _______________________________________________ >>> ceph-users mailing list >>> ceph-users@xxxxxxxxxxxxxx >>> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com >>> >> _______________________________________________ >> ceph-users mailing list >> ceph-users@xxxxxxxxxxxxxx >> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |