thinking sysadmin

While I use VMware’s free products - VMware Server and Player - on a daily basis, I’m very much given to wondering whether VMware has a future. As a company, they seem to be surrounded on all sides - Xen and Microsoft’s Viridian are closing in fast on one flank (well, at least Xen is - but I’m sure Microsoft isn’t going to let this opportunity pass by), and I wonder if Solaris Containers, Virtuozzo/OpenVZ and technology like FreeBSD Jails have escaped their blind spot on the other side.

At work, the role of VMware Server is already diminished to the case where you need to run more than one operating system on a given piece of hardware. Elsewhere, Containers, OpenVZ and FreeBSD Jails give us far greater density, better performance and better manageability. VMware gives us… well, a pretty stable free product that - subjectively - doesn’t seem like it’s doing us any favors on I/O performance, although it gets by. We looked at their for-pay product line, but the gulf between what was offered and what was charged was just too great.

OpenVZ - an Open Source offering I’ve grown quite fond of - gives us better density (11 moderate-use virtuals with plenty of breathing room on an older Dell PowerEdge 1650 with only 2GB of RAM is my best example), better deployment tools, and better management tools. In particular, VE creation speed is stellar, and vzyum makes managing updates for virtuals simple. OpenVZ doesn’t eliminate the need for tools like Cfengine, but it does work nicely in tandem with them. VMware doesn’t have an answer for this in our environment.

(We’ve also found that FreeBSD jails and Solaris Containers perform comparably to OpenVZ in terms of density and speed, although OpenVZ seems to have a few more features.)

I’ve heard it argued that using container-based virtualization increases the number of technologies that you need to learn, and thereby increases the chances for administrator error, since you use different tools for each OS. Additionally, the argument has been made that you lose density with the container approach, since you need a separate server for each OS. Even excepting technologies like BrandZ, I see both of these views as misleading: If you’re deploying OpenVZ, it’s likely because you already are running Linux and need to run more of it - the learning curve for the additional commands needed to manage the virtual is very shallow. The density argument only holds water if you have just a few virtuals and don’t expect to outdo VMware’s abysmal density. Can’t fill a server with just OpenVZ VEs? Run OpenVZ under VMware if you must: We’re using it in production, and it works quite nicely.

I will concede that VMware’s high availability tools today look like they work very well - and I say “look” because we can’t afford them and don’t use them today. Justifying the extra capital and operational costs of failover hardware doesn’t fit into the way we operate (work is an academic non-profit); recovery from backup is “good enough” for those applications we virtualize today. Even giving VMware the nod here, it’s worth noting that OpenVZ does offer instructions on how to build a HA cluster with DRBD and Heartbeat (with no SAN/shared storage needed - not something I see EMC-owned VMware too likely to promote, but I could be wrong).

So where do I see VMware going? I see this as the high point of the technology’s lifecycle. They’ll probably become entrenched in the high-end, helped along by EMC. For those of us that find EMC a little too spendy for our budgets, they’ll stick around on legacy machines if they’re in the shop already, and as time rolls on, there will be fewer and fewer new VMware deployments, with tools like OpenVZ being used in VMware’s stead.