Slow to take snapshot with guest memory - 6MB/s - mainMem.partialLazySave

[jsmurray]

Although I don't usually take snapshots with guest memory, I notice that doing so appears to be very slow averaging 6MB/s to the datastore.

As the 4GB of RAM is safely stored away, the throughput is very steady at betwen 5 and 7 MB/s with around 10ms latencies and I can't explain this.

Other transactions are normal, for example writing a large file within the same VM results in approx 60MB/s write to the datastore.

Storage vMotion between datastores with differing block levels occur between 60 and 80MB/s.

It's only when I take a snapshot with guest memory that throughput is low.  There is nothing else going on at the time on the host or the SAN.

I'm running ESXi 4.1.0 buld 433742 with 2 x 1Gbps NICs using Round Robin and iSCSI best practice to connect to a pair of Cisco 3750G switches and on to the SAN.

Load is perfectly balanced between my vmnics during this at around 3.0MB/s each.

Taking a 4GB snapshot takes approx 11 minutes to complete, and a vm with 24GB RAM takes over an hour.

If I shutdown the VM and create the Advanced Config Parameter  mainMem.partialLazySave = false I see a huge improvement.  4GB snapshots then take just 2 minutes and throughput is a far more healthy 50MB/s.

I can't find much info regarding the mainMem.partialLazySave setting or the way in which snapshot writes are slowed down.  Can anyone explain this and help find a way to not have to set this parameter to get decent speeds?

[chriswahl]

I can't find much info regarding the mainMem.partialLazySave setting or the way in which snapshot writes are slowed down.

I think the official stance from VMware is not to document any of the advanced vmx configs. I typically resort to trial/test or check sanbarrow.

http://sanbarrow.com/vmx.html

As for mainMem.partialLazySave, it determines if the snapshot will be performed as a background task (true) or not (false).

[jsmurray]

Yes, that's where I found the setting, plus lots of VMware Workstation and VMware Server threads regarding performance tuning (although I guess I did ask what the setting did, so erm, thanks).  I guess I should have asked how it works.

What I'm struggling with though why this is affecting my VM's.  There is very little IO going on, so why should this particular cluster be throttling back it's writes so much by default?  I cannot get it to write the snapshot faster without this setting.