09-14-2014 09:41 AM
We have had Fabric Watch for many years. I have been looking at the port fencing feature. It looks like the solution to single connections causing SAN wide latency, but have been scared by the possibility that under the wrong circumstances it could potentially disable both SAN A and SAN B connections to a single server, or potentially disable all ISLs to an edge switch. Are there some safeguards for this? Also does a fencing event trigger SNMP alerts - if it was to get fired off then I would want someone looking at it straight away.
09-15-2014 02:40 AM
i would not configure port fencing on a ISL port. On the other hand, yes, you can configure snmp traps/mails/events to be reported when a port is disabled by port fencing.
10-01-2014 06:30 PM
I would use Fencing on E-ports but just resticted to errors of a physical nature. (Encoding/Decoding, CRC, LO-SYNC, LO-SIG etc.) These are the errors that the switch cannot correct and will cause havoc in the fabric.
I would not use fencing for latency or congestion since the causes are most often somewhere else.
Using fencing on F-ports for latency purposes is a good idea. If a port hit a threshold and is shutdown it immediatly relieves the remaining traffic of credit issues. Fencing a port for congestion is asking for more trouble since the MPIO layers on hosts will just redirect the same traffic onto another path resulting in the same problem. Congestion is a result of either bad planning or lack of operational management and monitoring.
Hope this helps.