What ONTAP version?

There's been dramatic improvements in the deswizzler in 7.2. Also, do you snapmirror this volume regularly? One issue is that I think during early deswizzler phases at least is that snapmirror updates cause even the source volume's deswizzler to restart. (checkpointing of the destination deswizzler was fixed in 7.2.4, but not the source). You may want to suspend snapmirror updates until it finishes, or at least progresses to the next 'pass' (1/normal, 2/normal, 3/normal, then 0/inofile through 3/inofile). Use 'priv set advanced; wafl scan status' and note the scanid. If the scanid changes after a snapmirror update, you'll know that it's restarting. The scanner is supposed to run faster and catch up if it does restart, but I haven't found this to be much the case in practice.

You can raise the wafl scan speed to supposedly make the scanner go faster, but in my experience it never had much of an effect.

Unless you have a ton of deswizzlers running, trying to kill them off to force only a few to run at a time probably isn't worth it. It's probably the case that they are restarting when you don't want them to.

Also, the only (not quite safe) way I've found to force a single volume's deswizzle to resume is to quick do a 'vol restrict' then 'vol online'. The only other safe method is the recently added (7.2.4?) 'options wafl.deswizzle.enable off' then "on", but that kills/restarts everything.

Oh, also use 'wafl scan status' and note the snapid, and correlate it with the snapid in 'snap status'. (It deswizzles starting at the oldest snapshot). If it's stuck doing the oldest snapshot, you may want to give it a boost and trim back on your snapshots temporarily. If you have a lot of filesystem (especially inode) churn, deswizzles will take much longer. Note that snap 0 is the base filesystem.

--Dave

On Tue, Jul 15, 2008 at 1:19 AM, Kappe, Michael Michael.Kappe@izb.de wrote:

Hello toasters,

i have several volume deswizzling processes runing. This is ok, why i copy the volume with snapmirror from one aggr to another. This volume deswizzling processes now running for more than a month.

Now the question: When i stop the deswizzling with "wafl scan abort", can i restart them one by one to finish them a bit faster? What is the command?

Doing a reboot will restart deswizzling for all volumes. This is not an option ;-)

Regards

Michael

On Tue, Jul 15, 2008 at 2:02 PM, April Jenner aprilogi@yahoo.com wrote:

Hello toasters,

...

...

i have several volume deswizzling processes runing. This is ok, why i copy the volume with snapmirror from

one aggr to another.

...

This volume deswizzling processes now running for more

than a month.

This is normal and okay. If you aren't having any performance problems, then I'd allow it to continue.

I wouldn't be so quick to judge normality. A month for deswizzling is IMHO far far too long, and indicative of a either an extreme workload (lots of inode churn) or a system interaction causing the deswizzler to not progress. Both of these should be considered even if you don't see any clear performance problems. You may only see issues sporadically, or it may only be noticeable during certain kinds of workload. If nothing else, assuming bugs or issues aren't ruled out, it may be an indicator of an undersized head. I've found deswizzler issues to be fairly insidious.. since they tend to use a fair amount of RAM at times, and can have dramatic effects coupled with certain types of operations (such as ndmp backups), and the deswizzler itself can have odd behaviors (as previously mentioned).

--Dave

What version? There have been big improvements in 7.2 and dot releases.

...

...

Now the question: When i stop the deswizzling with "wafl scan

abort", can i restart them one by one to finish them a bit faster?

...

What is the command?

The big question here is whether or not you are having a performance problem caused by deswizzling. If performance is okay, I'd let it go.

Be careful of changing the values for wafl scan. You can cause problems by using the wrong settings. Changing the value for wafl scan effects more then deswizzling and in general is considered a bad idea.

If you believe you are having performance problems, it might be worth calling support and asking them help you evaluate your system. Other things might be a problem.

One last question, if you precieve a performance problem, do you have any measurements such as timing the time it takes to write a file or looking at perfstat output? Just curious.

--April

R200 - Ontap 7.2.4 Snapmirror is only used once - for copy the volumes The volumes are used for Snapvault - every nigth a new snapshot. The deswizzling process used sometimes up to 25% cpu.

The problem is that the R200 is not used only for backup, also for some fileservices. Also the monthly backup with TSM have poor performance on the volumes with deswizzling process running. Not very poor but annoying.

-----Ursprüngliche Nachricht----- Von: Dave Barr (鬼佬) [mailto:barr@google.com] Gesendet: Dienstag, 15. Juli 2008 21:45 An: Kappe, Michael Cc: toasters@mathworks.com Betreff: [Spam] Re: volume deswizzling

What ONTAP version?

There's been dramatic improvements in the deswizzler in 7.2. Also, do you snapmirror this volume regularly? One issue is that I think during early deswizzler phases at least is that snapmirror updates cause even the source volume's deswizzler to restart. (checkpointing of the destination deswizzler was fixed in 7.2.4, but not the source). You may want to suspend snapmirror updates until it finishes, or at least progresses to the next 'pass' (1/normal, 2/normal, 3/normal, then 0/inofile through 3/inofile). Use 'priv set advanced; wafl scan status' and note the scanid. If the scanid changes after a snapmirror update, you'll know that it's restarting. The scanner is supposed to run faster and catch up if it does restart, but I haven't found this to be much the case in practice.

You can raise the wafl scan speed to supposedly make the scanner go faster, but in my experience it never had much of an effect.

Unless you have a ton of deswizzlers running, trying to kill them off to force only a few to run at a time probably isn't worth it. It's probably the case that they are restarting when you don't want them to.

Also, the only (not quite safe) way I've found to force a single volume's deswizzle to resume is to quick do a 'vol restrict' then 'vol online'. The only other safe method is the recently added (7.2.4?) 'options wafl.deswizzle.enable off' then "on", but that kills/restarts everything.

Oh, also use 'wafl scan status' and note the snapid, and correlate it with the snapid in 'snap status'. (It deswizzles starting at the oldest snapshot). If it's stuck doing the oldest snapshot, you may want to give it a boost and trim back on your snapshots temporarily. If you have a lot of filesystem (especially inode) churn, deswizzles will take much longer. Note that snap 0 is the base filesystem.

--Dave

On Tue, Jul 15, 2008 at 1:19 AM, Kappe, Michael Michael.Kappe@izb.de wrote:

Hello toasters,

i have several volume deswizzling processes runing. This is ok, why i copy the volume with snapmirror from one aggr to another. This volume deswizzling processes now running for more than a month.

Now the question: When i stop the deswizzling with "wafl scan abort", can i restart them one by one to finish them a bit faster? What is the command?

Doing a reboot will restart deswizzling for all volumes. This is not an option ;-)

Regards

Michael