Start a Conversation

Unsolved

This post is more than 5 years old

745

November 9th, 2010 01:00

CX-300 metalun performance drop.

Hello,

on my CX-300 i have a Metalun with one host connceted to it.

It serves as a fileserver and the volume size is 1074 GB.

On the other server I'm making backup of this volume but not directly but through snapshots.

Recently I have noticed big drop in performance during backup.

Backup job rate have decreased significantly.

I have tried to run Diskeeper on the fileserver. It shows that the volume is heavily fragmented.

I have turned on automatic defragmentation, run manual but it didn't help me a lot.

Maybe the problem is that there is only about 17% of free space.

Any idea what might caused the drop of performance ?

Thanks.

131 Posts

November 9th, 2010 07:00

Do you have analyzer on your array?  If so, some graphs or a .nar file might help.  You can also submit a .nar file to support and they can help you.  I don't even think you need analyzer licensed for support to take a look - I think they can pull .nar files themselves without the license.

Did the performance drop happen suddenly?  Or has it gotten incrementally worse over a period of time?

Is yoru metaLUN a striped or concatenated metaLUN?

If it's striped, are you striping horizontally?  (component LUNs in different RAID Groups) or are you striping vertically? (component LUNs in the same RAID Group)

Are you using SnapView for your backup volume, or a Volume Shadow Copy?

If SnapView - Are your reserved LUNs in a different RAID Group than your production volume?  Are a lot of writes either going to the source LUN or the SnapShot during the backup window?

Also, host-side.  What is the file count of the backups?  How has it changed since performance dropped?  Lots of small files will add to your backup window - If someone dropped lots of small files in there recently (i.e. lots of small log files) - That can really affect your backup performance.

1 Rookie

 • 

20.4K Posts

November 9th, 2010 07:00

Navi analyzer did not get bundled on older CX arrays, you either purchase it or support can install “rebootless” Analyzer for troubleshooting purposes. On CX4 you can get support to pull down the naz file and decrypt it.

4 Posts

November 9th, 2010 08:00

Thanks for Your answer.

I have never used Analyzer and I'm rather sure I don't have it on the array.

The performance have dropped suddenly.

It's a striped MetaLUN. Component consist of LUNs from different RAID Groups of the same type RAID5 and size.

For backup I'm using SnapView. Reserved LUN pool is a different RAID Group consisting of small 50 GB LUN's allocated to source LUNs when needed.

I'm not sure what is more used during backup, source LUN or Snapshot but usually source MetaLUN is using just 50% of 2 allocated 50GB reserved Luns so

it seems that sorce LUN is much more used.

The host is a Fileserver so there is a lot small files. There are big files also but mainly there is lot of typical office files.

Backup have always suffered when processing small files but never so much like now. Number of files didn't changed significantly since it was ok.

Since few days I'm running defragmentation on that volume hoping it helps but so far no success.

4.5K Posts

November 9th, 2010 11:00

If you have updated your flare on the CX300 to flare 26, you now have Analyzer installed - most likely in the unlicensed mode - this means that you can start and stop analyzer to collect data, but the archives created will be encrypted - you will need support to read the archives. See Primus emc218359 for more information about using and configuring Analyzer.

If you're using snaps for backup, everytime you write to the source LUN it will copy that write (COFW) to the reserve LUN (it will copy a large block of data that contains the first write - all other writes to this same block will not be copied as the block has already been copied). When you run defrag you may be writing to a whole bunch of locations on the LUN and since all of those are writes, you could be touching all the locations - this will increase the amount of Reads from the reserve LUNs that you need to perform for a backup. Once defrag is complete, it should go back to normal.

Are you starting more than one snap on the LUN - say one snap for each day? Remember that the more active snaps on the LUN, the more IO you will generate for each Write to the source. If you have 7 active snaps, each write will need to be copied 7 times.

glen

4 Posts

November 10th, 2010 00:00

No I have flare 24 so no Analyzer. It's grayed out in the context menu.

is it possible for EMC support to analyze the performance or They will have to upgrade the flare ?

Normally on this LUN there are 4 active snapview sessions per day but it never was a problem.

I'll wait until defragmentation ends and then We'll see. It's going very slow so it could be even few days.

Thanks.

4.5K Posts

November 10th, 2010 12:00

Release 24 also contains the Analyzer program - please see emc163368 and emc255033.

glen

4 Posts

November 10th, 2010 15:00

Ok I have started the analyzer and I now I'm able to send it to support if needed.

Thanks fo help. 

No Events found!

Top