Hello everybody,
I am implementing a Scale-out file server together with Hyper-v, both clusters have been equipped with Chelsio nics which have been placed in different VLANs (to enable SMB multichannel). I ran some file copy between these systems in order to see how RDMA is working and I have observed some strange and inconsistent behaviours. I was expecting to see the same speed when moving files between systems that shares the same network infrastructure (all servers are connected to the same switches) and that mount the RDMA nics.
The network team implemented MLAG on the switches on which the RDMA nics are connected.
I just wondered if the implementation of MLAG could potentially interfere with the proper operation of the SMB 3.x ?
The behaviours that I have observed can be summarized as:
File copy between the Hv hosts - I manage to achieve a speed of 1.5 GB/s, but during this process the speed fall down (almost to 0 MB/s) for two times. it seems that I have reach a bottleneck or some hard limit.
File copy between the FS cluster node - I managed to achieve a speed of 600 MB/s, but at 1/3 of the copy the speed is reduced to 135 MB/s without been able to recover to the 600 MB/s.
File copy between the Hv hosts and FS - I managed to achieve a speed of 600 MB/s.
Thank you for the help J