Hi guys, we had some issues with our cephfs last, which probably have been caused by a MTU mismatch (partly at least). Scenario was the following:
OSD Servers: MTU 9000 on public and cluster network MON+MSD: MTU 1500 on public network CephFS Clients (Kernel Mout): MTU 9000 on public network RBD Clients (Openstack Nova): MTU 1500 on public network mostly (only one host had MTU 9000) We are running Jumbo Frames since 2 years and never had any issues with our RBD Clients, despite the MTU mismatch. Can someone explain to me, why this worked at all? From a network point of view, all osds should have used Jumbo Frames, which couldn’t be received by the most of the RBD Clients. The only explaination i can think of is that the ceph code catches the mismatch and doesn’t use Jumbo frames for Clients which don’t support them. Can someone verify this? Follow Up question would be: Is this mechanism not present in the cephfs kernel mount? (This would explain most of the problems we saw). Ceph Version is 16.2.10 across the board. Kernel used for cephfs kernel mounts is 5.4.0-135 Regards Felix --------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------- Forschungszentrum Juelich GmbH 52425 Juelich Sitz der Gesellschaft: Juelich Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498 Vorsitzender des Aufsichtsrats: MinDir Volker Rieke Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender), Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt, Dr. Astrid Lambrecht, Prof. Dr. Frauke Melchior --------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------- _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-le...@ceph.io
