Hi,
Recently I've been doing some performance changes for inter-lpar configs and completing some testing. So far test have been successful and promising but only completed in a dev environment.
Here is what I've done
created a separated vlan for inter-lpar communications separate from the VIO.
adjusted the MTU to 65280, and enabled Large Send.
transferred have increased by aprox 70% in some test. lowest was a 50% increase from the original configurations on other servers. I did see a improvement in terms of CPU as well compared to using default settings.
so my main question is this....
overall it was great. the problem which i was expecting was sftp/scp.
using nmon i did confirm they were transferring on the correct ports but transfer times were extremely slow.
I was using a 6GB file for test, on ftp best avg time was 14 seconds (all test were completed by a average of 10 test) when transferring 6 6GB files at the same time, avg time was 23 seconds.
Using sftp/scp transfer times were 2:30 min. 6 files approx 5 min.
so my next step was to try changing the algorithm on these processes which i personally was trying to avoid. I didn't run 6 concerent test with this one but the avg time was around 1:30 min.
Here was what i used for sftp:
sftp -B 260000 -o Ciphers=arcfour -R 512
here is what i used for scp:
scp -c arcfour
from ifconfig on this interface:
tcp_sendspace 262144 tcp_recvspace 262144
when using MTU=65280 and largesend what would be the best thing to adjust in terms of network tuning.
As I said this is just dev testing and I know real cases will be different but I'm just trying to see what improvements I can make since some people have been complaining about slow transfers between systems and this was the best i can think of at the moment.