We mirror two EVA 8100 with 70 disks each in default DGs and observe that as disks get replaced the numbers of allocated GBs on the two sides of the mirror drift hundreds of gigabytes apart and never level back until controllers in one or other EVA are re-booted, usually the more highly allocated side but occasionally the other side too has to be re-booted to force the two EVAs to show the exact same allocated size as should be expected due to the absolutely equal allocation policy on both sides of the mirror.
The firmware is 6220, the most up-to-date.
Is there a less intrusive way to flush the correct values of allocated space to the CommandView tool ?
Re-booting a controller has consequences for the client systems because hardware paths leading to a re-booting controller may need minutes to fail-over to an alternate path via a still running controller.
For a product in its 6th incarnation (F/W 6220) one would expect diagnostic tools to display reliable values without having to resort to crude arm-twisting, but HP seems loath to address the issue, probably because most users don't mirror their EVAs and are not even aware that something's wrong until the discrepancy causes the value of "Allocated" to exceed the value of "Total", in which case HP suggests to re-boot the system as a aquick fix.
So long as HP doesn't fix the root problem we are stuck with re-boots, but how to minimize the impact on the running clients ?
We can locate the respective controller for each hardware path but how to temporarily promote a still running path as preferable even though it is not the best one due to the asymetric nature of controller-disk interconnects ? Auto-path algorithms in HP-UX 11.31 might even actively obstruct such attempts at prioritizing.
Shorten the IO_TIMEOUT ? Sounds too risky but who knows ?
