A few weeks back a colleague of mine, Martijn wrote an article regarding Citrix Provisioning Services. Recently Citrix came across, by accident, a rather natty new feature where by the write cache could be homed in RAM but also spill over into HDD should it become saturated. Quite clever and obviously blisteringly fast if all the writes and reads sat in RAM all the time.
Some of the comments on his blog are really interesting and, as I found when I was at Citrix, the PVS vs MCS debate has many sides.
I think the point Martijn is trying to make is while there are many ways of doing things the majority of people are striving to do things better which could be less complexity, more integration to the application and systems that are easier to implement.
I was at Citrix long before Ardence was acquired and while it’s a great technology to deploy servers and VMs from it’s not one to implement unless you are 100% in control of it and all the other ancillary services it needs. If you are then great but the upgrade path was a real pain before and when it’s the core system keeping things working that can be a bitter pill to swallow. MCS was driven by a requirement for simplicity, actually by how easy Linked Clones were to push out. It used to take several days to get a small XD POC up and running and that was all down to PVS. MCS reduced that to a few hours because we were doing things the way customers understood. MCS has grown up a lot since the first FMA implementations too and scales just as well as PVS, with a lot less moving parts. It’s without complexity, integrated into XD and is overall a simple concept to adopt by customers.
PVS is non of the above. DHCP, networking, target config not 100% spot on? Potential issues. That’s before the PVS config even gets implemented.
To address the latest feature in PVS, while sticking everything RAM sounds good you have to be extremely accurate on how it’s implemented and know pretty well how things will change in the future. There’s also the question about the HDD that it still needs – place that back on a SAN and you’re back to square one, put it on a local HDD and no easy live migration, at least not out of the box*. A rather large issue if upper management have a session on the XA at risk after a hardware failure on a host.
Hoenstly I couldn’t care less whether someone uses PVS or MCS so long as it works but in my experience technology needs to be easy to adopt, change and grow. Stick PVS on a Nutanix cluster and we address some of those things. Put MCS on anything and it does the same. Put MCS on a Nutanix cluster and all of them go away.
There’s a place for both and the customer should decide which they want and a consultant (vendor or not) should be there to support and advise regardless of whether it benefits them or not.
*Thanks to Jarian Gibson for pointing out that you could do a vMotion/live migration with a local HDD through other means using this sort of thing. Additional steps we mitigate in PVS or MCS 🙂