At the start of September I began an experiment to track the disk growth of View Composer Desktop that was not recomposed or refreshed for a while. The initial article last month covered the first month of growth and wasn’t very surprising, an initial step of 1GB followed by sporadic growth of another 1 GB for a total of just over 2GB after a month. I left the VM running and taking critical updates, over the first two thirds of October the growth was steady, after seven weeks the delta had reached 2.5GB.
Today I noticed that Office 2007 SP3 had started to deploy through WSUS, so I took a look at the growth of the VM again. In the 3 days since my last check the delta had grown from 2.5GB to 4.7GB, eating about another three months worth of extrapolated growth.
This isn’t a big issue for me as I had only a single VM sitting on an NFS datastore with nearly 2TB of free space. But what if you had 200 or 2000 VMs and had only budgeted for deltas to grow to 4GB? When the update was applied you’d run out of space on every datastore holding affected desktop VMs and every affected desktop VM would stop, showing a question about out of datastore space. This is standard vSphere behavior when a datastore runs out of free space, every VM on the data store halts.
Every user would loose access to their desktop.
This is not something you ever want to happen, not for server VMs and not for desktops either.
So what does this mean?
- View Composer doesn’t produce totally predictable disk use. Like any thin provisioned storage you must budget and monitor available capacity carefully. The important thing is to understand the consequences of design decisions and accommodate them in operational processes.
- If you are going to use WSUS or any other mass patching tool to update deployed Linked Clone VMs be very careful about what patches you release. Test patches on a sample Linked Clone VM and measure growth before releasing to production.
- Better still, use the recompose feature to deploy updates to Linked Clone VMs. Refresh and recompose regularly. Put the Linked Clone VMs in an AD OU that doesn’t receive automated updates.
As always there are multiple possible solutions and no solution is universally applicable. know your options, accommodate the compromises that must be made to accommodate the requirements and document everything, especially the why.
© 2011, Alastair. All rights reserved.