I had planned on migrating several of my SQL server LUNs to larger LUNs this weekend. So as a good admin and since I've never done it before, I wanted to test on a non essential machine. I create a new LUN, 50G, assigned it to the server, it shows up as expected. I copied data to the LUN, just some basic stuff. I then create a new larger LUN, 100G and then did the migration. In NavisSphere it shows that the LUN is migrating, but when I look at migration summary, it says 8 hours. Also the rate is on High..
Should this really take 8 hours? Is there anything I can do to speed it up? Can I cancel in the middle? Did I do something wrong?
Also, and this may just be a coincidence, but I got reports that several of our Apps were having time out issues right after I started the migration.Since all those apps talk to the back end SQL DBs, which is connected to the SAN.I’m wondering if they are related and if so how do I prevent that this weekend, or if not how do I prove that so I can verify doing the migration is not a bad thing.
is your target LUN on the same raid group as your source LUN ? If yes then you are doing a lot of thrashing which could also impact other applications on the same raid group. You can always abort LUN migration at any time without any issues. Expanding could give you more performance (striped MetaLUN, don't stripe LUNs in the same raid group) as you could spread your workload over more spindles but it takes much longer as data has to be restriped and the new space not available until restriping is finished.
The calculation for estimating the duration of a migration can be found in the EMC CLARiiON Best Practices for Performance and Availability, FLARE Revision 26 (CLARiiON Best Practices). (Search the paper for 'LUN Migration'.) Note this is CX4 CX3 document, so the rates will be marginally quicker than your CX3. Also, as previously mentioned, the provisioning of the source and destination LUNs will affect the rate. Best Practices includes a discussion of the affects.
CLARiiON Best Practices is available on Powerlink.
Thanks for the replies.. My target and destination LUNs are not using the same raid group. When I created the source LUN it was using raid group 9 which is comprised of disks 0-6 in DAE 3 and the target LUN was using raid group 10 which was comprised of disks 7-12 in DAE 3.
Maybe because they were in the same enclosure? Maybe my raid groups are to large and Raid 5 over 6 disk is just taxing in general..
I also found something else to be odd... In my server it doesn't automatically show that the partition grew by 50gb. If I look in storage manager I see the extra space but its setup as unallocated space. I'm thinking I still have to use diskpart here, but didn't think that was the case with migrations.
nothing jumps out as far as your raid group are concerned, maybe there were already taxed pretty hard and LUN migration pushed them even harder ? I let my migrations go at Medium speed, don't really care how long it takes as long as it completes ..sometimes i can crank them up to High on weekends.
yes, you will need to extend the partition into the new space.
These are brand new raid groups, the disks assoicated to them are brand new as well and not allocated to any storage group (besides my one test server), I had only created those two LUNs. Again though it could have been a coincidence that the Apps starting acting up, actually come to find out it may have only been one App, leading me to believe it wasn't the SAN. I should check the SAN resouces though as I'm doing this next time so I can see if its some how effecting the overall proformance. I guess that's my next post.. :-)
Btw, you indicated that I could stop a migration while its in progress, with out any adverse effects to the source LUN and its data. However I didn't happen to see that as an option anywhere, did I overlook it? Can you point me in the right direction?
finally...I did do the diskpart and the partiton expanded as expected and my test data was still there..
If you go to the LUN Properties window of the source LUN, you will see a tab called 'Migration'. That will show the details about the migration process and also allows you to cancel the migration process if you want to.
LUN migrations can affect performance; yes I have brough down entire production apps as a result -- newbie years -- Since them I have implemented a best practice methodology for when I do them:
Migrate at Medium or lower. -- ASAP and High can cause contention at the following points depending on current utilization:
SP - Especially if the Migrating LUN and Source LUN are on the same SP.
BUS - Especially if the Migrating LUN and Source LUN are on the same BUS -- same goes for DAE.
Raid Group - Never target and source the same RG.
Estimate (at Medium) 1 Hour for every 10GB for Migration/Expansion (use the target total size).
Ensure FAST Cache is OFF on the LUN being migrated/expanded.
If your estimates show that your migration will go into prod hours, you are better off letting it go into prod hours on LOW and extend the outage time rather than Medium or High and risk performance impact.
Bob, please remember to mark your question as answered when you get the answer or solution that fits your issue. This helps when others are searching for solutions for similar issues. Also, award points to those answers that provided the most correct answer.
Thanks for the tips.. We did get through one of the migrations this past weekend, the process took 24 hrs on High, but we saw no impact to our production Apps, well no one complained anyway. We've got one more to go, and I'll use your info handy for that one.
dynamox
11 Legend
•
20419 Posts
•
87439 Points
820
1
Posted August 17th, 2011 21:00
is your target LUN on the same raid group as your source LUN ? If yes then you are doing a lot of thrashing which could also impact other applications on the same raid group. You can always abort LUN migration at any time without any issues. Expanding could give you more performance (striped MetaLUN, don't stripe LUNs in the same raid group) as you could spread your workload over more spindles but it takes much longer as data has to be restriped and the new space not available until restriping is finished.