We have two data centers that used to have a layer 2 point-to-point with a 1518 MTU. There are servers at each site that do iSCSI replication back and forth. The interfaces on the servers set their MTUs to 1500. (I unfortunately don't have access to them and cannot tell yet whether Path MTU Discovery is enabled).
We have migrated to SD WAN and now the same circuit is used between our Silver Peak appliances at each data center. Now data-center interconnectivity is via 3 underlays: The original p2p, and 2 Internet connections. However with the encrypted tunnel overhead, the MTU is 1488 over the WAN.
So the replication used to occur over the layer 2 point to point with 1 ms latency and 1512 MTU.
Now it's going over the SD-WAN - specifically in an overlay tunnel that has 10 ms latency. The server admin has mentioned that the replication takes longer than it used to.
Hypothetically, if the Servers are sending 1500 byte frames and don't have Path MTU Discovery - would they benefit more from switching to an overlay that has lower latency (sub 1 ms), or if the servers were to enable Path MTU Discovery and reduce the frame size?
No comments:
Post a Comment