We're looking to improve throughput between VNX's over a WAN with about 21ms RTT. It looks like the BDP would be around 2.9MB over this 1Gb link, and I know we have default settings for tcp sndcwnd on both sides as well as the default for tcpwindowlowat (RCP). So my question is if this is a one-way replication, should the tuning be done on the source datamover or both?
I can see in a capture the max window size they negotiate is just over 1MB already, but, it's not a lossless network of course, so the intent is to just get a bit more out of the link in order to meet the RPO. FastRTO is not on (yet).
We have to avoid unnecessary reboots of these to a bare minimum so I'm avoiding a bunch of "tests" trying to make one config change and hopefully get closer to the theoretical max of that link.
Between a VNX2 and VNX1, but the tunable parameters are the same, correct?
What I'm looking for is more detail around those settings and whether it mainly applies to the client(source) side or if both need to have the same settings to get more throughput.
DELL-Sam L
Community Manager
•
8150 Posts
•
34715 Points
1144
0
Posted May 14th, 2020 10:00
Hello downhill2,
Which VNX system do you have? Which version of replicator are you running currently?
Please let us know if you have any other questions.