Page 1 of 1

AO new data written to wrong CPG

Posted: Fri Dec 20, 2013 2:53 pm
by skumflum
I’ve setup AO 3-TIER. The Volume user CPG is FC disk but new data is written to NL disk. If I run statpd from CLI when I storage vMotion a VM from our EVA to 3PAR only NL disk generate IO.

When I tested functionality on the volume it worked initially as expected and all data first was placed on FC disk. I created some big VMs, generated IO on some drives and created stale data on others. OA moved data to the other CPG as I would expect.

I then deleted the VMs and emptied the VMFS LUN. Now new data is written on NL

How can that be?

Re: AO new data written to wrong CPG

Posted: Sat Dec 21, 2013 2:26 pm
by skumflum
I’ve run some experiments.

I emptied the volume completely and ran the reclaim tool from ESXi “vmkfstools -y 99”. I then waited for the 3PAR to get thin. My thoughts was that it maybe filled data back to the NL CPG because it was previously occupied by data from my previous experiments. It did not help and new data is still written to NL.

I then created a new volume with the exact same settings. New data is written to FC CPG like it should be.
Check out the screenshot.

“User Space Provisioning” for this volume is the FC CPG but “user Space Allocation” is RAID6 NL. What does that means?
The new volume I created with the exact same setting, the “User Space Allocation” is RAID5 FC.
I’m fully aware that I maybe misunderstand how things is supposed to work, I’m still 3par newbie. But ought new data not ALWAYS be written to the user CPG??

Re: AO new data written to wrong CPG

Posted: Thu Dec 26, 2013 7:22 am
by skumflum
Okay, there may seems to be some conceptual misunderstandings from my part! :oops:

If a region is occupied by data and the data is deleted on higher level (I deleted all the VM from the vSphere Datastore before testing), new data is then perceived as modification to the regions created from the VV. Therefore "new" data could be written to NL as long as the region on NL disks exists.

I did not wait long enough after I zero filled the Datastore and therefore regions in the NL disk still existed.

This is a lesson learned for me

Please correct me if I'm wrong :)

Re: AO new data written to wrong CPG

Posted: Tue Jan 07, 2014 1:14 pm
by Richard Siemers
Interesting discovery, thanks for sharing.

So, to properly reset your test, after zeroing, how long should someone wait? My guess is until the next AO cycle runs?