Hello all,
Our shop has recently migrated over to two Force10 S4048 switches into our Compellent SAN environment. Both switches are 48 port 10GB. Our SAN has three physical servers that all connect up to the two switches. The switches are doing VLT so we can do Port-Channels from the physical hosts to both switches. The switches are configured with different VLANs to segment off different types of traffic - user data, live migration, iSCSI, etc.
During our outage window we were able to pull everything up and all looked great. The next day once a bit more load was put on the SAN we found massive performance degradation. After some troubleshooting we were able to narrow the issues down to the iSCSI VLANs. Eventually after not finding a solution we moved all iSCSI links to two separate layer 2 10GB switches. Once we did this almost immediately all of our systems began performing as they should.
On both the switches and servers we are doing jumbo frames, no flow control. During the event the only oddity I could find on the switches were numerous input throttles on the iSCSI interfaces in question. Once this was found we ensured Flow Control was turned off on all hosts and the switches again. We are told by Compellent's support that the SAN does not send out pause frames, but I was under the impression that even if it was the switches would not honor them anyway as long as flowcontrol rx was turned off. Is this true? And if it is not, is there a way to ensure these pause frames are not being accepted? Is there anything that would be increasing the throttles counter on an interface that is NOT flow control pause frames?
We're kind of at a loss here. During troubleshooting this TAC had me try turning on flowing control to see if the issue would subside, but this had zero affect on the SAN performance.
thank you in advance all!
