VMware Cloud Community
groobo
Contributor
Contributor

Location of VM Pagefile

Currently our infrastructure is running on a netapp fas2020 filer. A volume was created specifically for the intention of holding temp files and page files. We called this ScratchVol and created a ScratchLun on that volume. We have a 3 MB link between our fas2020 and our DR site FAS 2020 filer. Utilizing Snapmirror we replicate several core volumes across this link. We created ScratchVol with the intention of omitting it from snapmirror so it would not congest the bandwidth as Snapmirror replicates deltas. We are now at the point where we need to increase the size of this volume to allocate enough space to new VM's and it is proving difficult as I learned from a VMware rep today that adding an extent to a LUN only works 50% of the time while the machines are on. (I noticed it doesn't work for me 100% of the time) She suggested I shutdown all VM's on the volume (ie. All of them as they all have contents on ScratchVol) and then attempt to add the extent. This is a maintenance nightmare. I guess what I am looking for is insight into other users infrastructures with regard to the above issue. At what point does managing page files on a separate volume outweigh the benefit from lack of saturating the link? What are you guys doing? Do you separate the page files for snapmirror, or a similar utility to replicate volumes across a link? Thanks for any insight.

0 Kudos
3 Replies
Rodos
Expert
Expert

Great question, and one that may be relevant to more people as they start to use SRM.

Mike Laverick who wrote the book on SRM does not cover this, however he is giving it some thought. Not that I speak for the council of the great wise ones.

Moving pages files to a separate disk is not an un-common practice. Windows figures it out if the disk for the page file disappears and recreates it on C. SRM currently has issues with this though, as it wants the datastore replicated (I need to get back into the lab and do some more testing on trying to get this to work).

But this is all digression. To your problem.

Why don'y you just create a new datastore and put the new machines page files onto that one? Sure one is tidy but ... Or create a new large lun and storage vmotion all the disks across with a script. How much bandwidth are you saving? Can you monitor the IO's on that LUN? If you get your memory settings right the machines may not page a lot and it may be okay to not separate the page files, only analysis of your environment will give you the answer.

As to your real question, are people separating page files for mirroring, yes, a quick search on the internet will reveal that traditionally many have been, but its a different question if its right for your environment.

Would be keen to hear what IO rates you are getting on the page datastore.

Considering awarding points if this is of use

Rodos {size:10px}{color:gray}Consider the use of the helpful or correct buttons to award points. Blog: http://rodos.haywood.org/{color}{size}
0 Kudos
groobo
Contributor
Contributor

How can I see the I/O rate of a single volume? It spans across all VM's on two different ESX hosts so I am not sure how to collect the data.

0 Kudos
groobo
Contributor
Contributor

Okay so I ran a lun stat command every second for 60 seconds at around 9 AM and attached the results. The total usage, although I am not sure this will be too helpful:

/vol/ScratchVol/ScratchLun (320 days, 20 hours, 40 minutes, 5 seconds)

Read (kbytes)   Write (kbytes)  Read Ops Write Ops  Other Ops  QFulls Partner Ops Partner KBytes

3820251618      1525467296      19248151 65879513   375955     0      0           0

Message was edited by: Ken.Cline to better format the output

0 Kudos