> > I gave up on this topic.. ceph does not properly support it. Even though > it > seems really promising. > > Tested a ping on 40gbit with rdma ud which took for both ways 6us. > > rdma rd i didnt got running (maybe my cards need some special > threatment..). > > tcp-ping at mtu 9000 took 20us > tcp-ping at mtu 1500 took 36us > > all measured against 4k blocksize. > > so the transport is up to ~3-4 times faster by using rdma. > How many % is that of a total read/write operation? Is there some sort overview what parts in a read/write action have what latency? _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |