I have two io domains and have passed a FC LUN to a guest LDOM via both io domains using the LDOM multipathing feature ("mpgroup"). When one io domain is shut down, failing over the io to the other io domain works without a problem.
But when the io domain is up again, the disk io continues to run over the io domain that remained online. This means that after a reboot of an io domain all disk io from all guest LDOMs (which use LDOM multipathing) continues to run over the other io domain. No traffic will be running over the io domain that was rebooted.
Of course every io domain should be sized big enough to carry all io in the case of a failure of the other io domain. But still the question: Is there a way to redistribute LDOM multipathing disk io more evenly when an io domain recovers and is up again?
- LDoms Manager software 188.8.131.52
- Solaris 11 on both io domains (pkg://email@example.com,5.11-0.175.0.0.0.2.1:20111019T075711Z)
Edited by: user6353652 on 25.05.2012 00:56
Edited by: user6353652 on 25.05.2012 00:58
Edited by: user6353652 on 25.05.2012 00:59
As far as I know then only way to "re-balance" the I/O across the domains is to unbind/bind the guests. Not a great answer, but this could be done as part of the guests patching cycle.
I think there is an RFE to provide MPxIO-like features to guests.