Put your pagefile on it’s own disk

If you’ve been in IT for a while, you may stick by the rule that with Windows, you should put your pagefile on it’s own disk. Originally, this was when computers had little RAM, and to optimise performance of the pagefile it was advised to use FAT instead of incurring the overhead of a journaled filesystem such as NTFS. Of course, much of this advice has become less valid with the decrease in RAM cost, and the use of shared storage with virtualisation means that the prevalence of dedicated pagefile disks is decreasing.

But, here are some reasons why, in vSphere environments, that you should have a dedicated partition for the Windows pagefile, even with Windows Server 2012.

Reasons to put your pagefile on it’s own disk

1. Disk I/O can be more controlled by SIOC
2. Using vSphere Replication, you can exclude your pagefile disk from being kept up to date, reducing network overhead
   • With SRM, when used in conjunction with vSphere Replication, you can exclude the pagefile disk from having changes replicated
3. Virtual machine disks can be on different datastores
   • You can leverage different performance characteristics of SAN capabilities, such as RAID 10, SSD, or even completely different SAN controllers
4. For high demand applications and systems, you can define a new virtual SCSI controller and get it’s own Queue

Your mileage may vary for each of these, but I will focus on what I consider to be the most beneficial reason to create a dedicated disk for your pagefile – for replication.

Creating a non-replicated pagefile disk

When Windows is running, there is disk I/O to the pagefile. When the available RAM is low and Windows needs to move less-frequently accessed memory pages to disk, this I/O increases. However, when Windows reboots, it disregards the contents of the pagefile.

When a VM is replicated by vSphere Replication, or SAN replication, the destination VM object is updated regularly with any changes to the disks – and if the pagefile is part of this replication scheme, then the I/O to the pagefile is also replicated.

So, you should create a new disk in each VM, and then move the Windows pagefile to it, then stop this disk from being replicated.

There is some advice on doing this at http://kb.vmware.com/kb/2009324 – however I believe this to be overly complicated and ‘technical’. This is how I would recommend you do it.

1. Set up replication for your VM, with all disks (including your pagefile / temp disk)
2. Wait for full replication to complete.
3. Reconfigure replication of the VM to exclude the pagefile disk.

This means that if you need to recover the replicated VM to a running machine, when it boots up from the replicated data, the pagefile disk is still there. One of the first steps that Windows does when it starts is to wipe the pagefile, so it makes absolutely no difference to Windows if the disk’s data is weeks, months or years out of date.