HPE EVA Storage

Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

 
SOLVED
Go to solution
Adam Garsha
Valued Contributor

Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

I am encountering a problem where demand allocated snapshots grow to nearly as big as the data on the source luns within only a period of a few hours. It is as if every darn data block is getting dirtied.

Has anyone seen this?

I've run sanity checks against Oracle data Luns, etc. and I do NOT see similar behavior.

Further, the log luns are NOT experiencing similar copy-on-write growth.

What gives with Exchange 2007?

I did have the exchange administrator disable database maintenance last night (since I thought maybe that was it... but no, still seeing the symptoms.)

Here is an example SSSU command demonstrating how I am obtaining the snap.

ADD SNAPSHOT snap-c1_emarqstaclus_lun021 VDISK="\Virtual Disks\windows\emarqstaclus\STA1-SG1-DATA\c1_emarqstaclus_lun021" ALLOCATION_POLICY="Demand" REDUNDANCY="Vraid0"

This thread is NOT to discuss application queisce, etc., but rather to understand why I am seeing so much copy-on-write with EVA demand allocated snapshots and exchange 2007.

I have:

EVA 8.1K XCS 6.2
Windows 2003 SP2, Exchange 2007 SP 1

Any ideas welcome about why so many dirty blocks. I've run out of ideas.
8 REPLIES 8
Adam Garsha
Valued Contributor

Re: Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

Hmm... "online compaction"... anyone know anything about this feature?
Uwe Zessin
Honored Contributor
Solution

Re: Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

The Microsoft knowledgebase comes up with a number of articles of you ask for 'exchange,online,compaction'. I am not an expert on Exchange so I don't even try to sort out any URLs for you - an Exchange administrator is better qualified for this job ;-)

I, too, guess that some activity is making massive changes to the database. Maybe the Exchange administrator did not disable all maintenance tasks, or something external is causing this.
.
Adam Garsha
Valued Contributor

Re: Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

One correction: exchange 2007 with pre-release SP2.

Yes, I've asked the exchange admin to look also and stressed I would probably get no where since this has to be on host (based on my tests and fact that log luns don't show similar activity).

I googled yesterday for hours, but I found nothing but meantions of "database maintenance". I am pretty confident that those were indeed not running for my tests.

Other: We don't use checksum scanning, nor page zeroing. Only interesting feature that I know is enabled is "online compaction" , but looking around that is the default.

So... I was hoping another SAN guy had hit this before since exchange is such a popular app. Sounds like I am alone here though or maybe one of few who wants to do demand allocated snapshot.

I am not familiar with MS knowledge base, but am eager to try out some searches and see if I hit something. I'll give it a try.

I'll let this question ferment a little then swing around with points.
sgparker
Occasional Advisor

Re: Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

Greetings,
Was more information ever found for this issue?
I see the same type snap growth for a large medical database on a 8000/6.2

Thanks.
Adam Garsha
Valued Contributor

Re: Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

I gave up on using snapshots since they weren't saving me much space and I am using a combo of ca dr_group/suspend and multi-mirrorclone with intermittent snapclone to get my copies.

I was able to get money to buy enough disk to make this work and we are already moving to next version of exchange were we will use DAG groups to get the copies at the app level.

For my situation, I think the problem was a form of online defragmentation of the b+ trees, but I never got anyone to confirm it that knows how the code works.

I am confident it was something the app was doing and not an EVA problem. I proved this to myself by snapping oracle db's and not seeing similar issues.
Adam Garsha
Valued Contributor

Re: Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

Curious, is the database that you are working against an intersystems cache database?
sgparker
Occasional Advisor

Re: Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

Yes, It is a Cache DB, approx. 1.5TB in size, I'm using clones also, lucky I have the disk available.
Adam Garsha
Valued Contributor

Re: Demand Allocated Snapshot Growth (EVA) and Exchange 2007 DATA luns

If memory serves from back in my Epic days, Cache DB's are b+ trees as well. Maybe that is the nature of the beast with this type of DB file (since exchange DB's are b+ also). Don't think so or can't remember, but maybe.

It is certainly true if there is any defrag of filesystem or DB itself that you can see churning that makes blocks look dirty to EVA.

Anyhow, good luck to you.