We noticed this degraded write performance too recently when the nearfull flag is present (cephfs kernel client, kernel 4.19.154). Appears to be due to forced synchronous writes when nearfull. https://github.com/ceph/ceph-client/blob/558b4510f622a3d96cf9d95050a04e7793d343c7/fs/ceph/file.c#L1837-L1839 https://tracker.ceph.com/issues/49406 >It might be more accurate to say that the default nearfull is 85% for >that reason, among others. Raising it will probably not get you enough >storage to be worth the hassle. > >On Tue, Apr 13, 2021 at 7:18 AM zp_8483 <zp_8483@xxxxxxx> wrote: >> >> Backend: >> >> XFS for the filestore back-end. >> >> >> In our testing, we found the performance decreases when cluster usage exceed default nearfull ratio(85%), is it by design? >> >> >> >> _______________________________________________ >> ceph-users mailing list -- ceph-users@xxxxxxx >> To unsubscribe send an email to ceph-users-leave@xxxxxxx >_______________________________________________ >ceph-users mailing list -- ceph-users@xxxxxxx >To unsubscribe send an email to ceph-users-leave@xxxxxxx _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |