11-26-2009 06:54 AM
I'm not referring to separate zones for disk and tape etc, but rather the specific switch port settings that maybe we should be paying close attention to when attaching native fibrechannel drives to FC switches.
Does such a generic set of best practices exist ?
In my case I'm connecting both IBM 3592s and a Dell ML6020 library into my ED-48000 cores.
Years ago an engineer told me that the following options seemed to work well:
"Lock the switch port's speed; set it to G port and disable E port functionality; disable NPIV"
And in case of the IBM drives, "Set the drives to the correct speed and lock them to F Fabric Node port."
Apparently, once locked as a G_Port, the switch attempts to initialize that port as an F_Port only and does not attempt loop initialization (FL_Port) on the port.
Q. Does this mean that tape and FL_Port is a no-no ?
Then he said that because a port designated as a G_Port can become an E_Port, this capability should also be explicitly blocked
Finally, tape might misbehave with auto-negotiated port speed, so lock the port to e.g. 4GB
All the above seems sound advice as the drives certainly work
But am I really getting the best from my infrastructure with the settings ?
Google searches etc haven't been that helpful because it's actually quite a difficult question to frame without generating thousands of irrelevant hits..!
Many thanks for any advice / pointers you might be able to provide.
11-26-2009 08:52 AM
Some drives (I think we found IBM LTO-2) look for FL first, so if you do nothing they come up in loop mode. The LTO-1 only supported loop mode, but the LTO-2 supported either.
We used portcfggport on all tape-facing ports for these, so the switch port skips past trying to connect as a loop. The drives then come up in full fabric mode.
Does it make a difference to anything? I'm not sure. If there were other devices on the loop it would, but these are point-to-point connections.
We don't do anything else at the switch. There might be a lot you need to do at the server! Typical server issues are multipath disk array software setting all HBA ports to suit the disks, e.g. with very short timeouts so path failover is quick, and large IO request queue sizes. Tapes need long timeouts so the drive can reset, and often quite modest request queues. IBM publish some (rather dated but likely still true) settings.