Correct in the "My HACMP didn´t detect errors," as HACMP does not need to do anything in that case. This is purely AIX LVM. You would have the EXACT same results w/o HACMP in this scenario. I
I have seen in testing i/o hangs to the primary/only copy left in the 3-5 minute range before. The PowerHA 6.1 Enterprise Edition redbook actually documented results of:
"The status of the resource group pokrg is still available in node Zhifa for around 5 minutes, and during that time the application appears to be hung and the users cannot write or read to the disks."
I would like to think there are some tuning parameters to help, but can't say I've had luck. I tried fast_fail on the fc adapter, hcheck_interval, and queue_depth and results were still about 90% the same.
My inclination, is its more fiber related. I have a long history with LVM mirroring and I don't recall seeing these significant delays in SCSI and SSA storage days. But I MAY have selective memory these days.
I would be greatly curious if support does give you some options that help this as I would like make note of it and push it out in our pubs if possible.