The hidden benefits behind the hidden costs of virtualization

A couple days ago I posted a link to an article on SearchServerVirtualization.com about the hidden costs of virtualization. I thought it made a lot of good points, and figured I’d take a little different angle to look at some of the less obvious benefits that come from the issues raised in the article.

Power and heat

The article points out that as you consolidate servers onto machines that run at higher utilization, you increase the power draw of a given rack and that may exceed the amount of power being supplied to that rack. This seems like a valid point, so let’s look at how we can mitigate this problem a bit.

Hidden benefit:

  1. New generations of hardware are putting a much greater focus on lowered electrical power consumption. Intel’s Core architecture in particular has really reduced the power consumption of the Intel line of CPUs. Typically, when you consolidate you’re pulling lots of little machines onto a few larger machines. 1U servers exist to serve the idea of one app per machine. With virtualization, you’re achieving the same goals but in a much more flexible manner. By using fewer, larger machines, you’re eliminating a large amount of wasted power draw required to simply keep the smaller boxes running, even when they’re idle a dual Xeon (non-core arch base) 1U server will draw over 200 watts. So for consolidation projects you want to look into larger machines than would have been considered in the past. Using a rack full of 1U or 2U servers and running them at 80% utilization is a problem, a rack full of 4U servers running at 80% utilization may or may not exceed the power available, but it will put you in a better position and combined with the next point the issue is truly mitigated.
  2. If you’re doing server consolidation then it means you already have the rack space for a larger number of servers. So use it. Don’t compress everything into a few racks, just spread the servers out more to balance the incoming power and cooling resources. It won’t save you as much square footage, but it will avoid costly rewiring of power trunks and potential hot spots where cooling isn’t sufficient.

Patching

The issue here is that you now have a whole bunch of applications running on a single physical machine so having to reboot that physical machine is very costly. Certainly an issue, but I’m not sure this will continue to be such a big problem going forward. There are two factors for this.

Hidden benefit:

  1. It’s looking quite likely that the base OS running on the real hardware will cease to be a full blown operating system. From the article it sounds like they’re thinking that Windows or something is running the virtualization environment. That is something I suspect will cease to be the case. VMWare ESX and Xen Enterprise have already gone the path of a minimal base OS that does nothing more than manage other virtual machines. If you’re running Windows, it should run within a virtual machine, not on the raw hardware. In this scenario, patch management should only be slightly more complicated because of the second point.
  2. Both VMWare ESX and Xen Enterprise support the ability to migrate running servers between physical host machines. This is a huge benefit of virtualization and as those technologies mature, patching the base OS will require only milliseconds of downtime for the applications. All you do is migrate your machines off the hardware, patch it, reboot and then migrate them back. Of course the fact that the base OS is a highly stripped down OS 100% focused on virtualization should also greatly reduce the number of patches required. The idea that you run a full blown OS like Windows 2003 on the server is one that will surely die out over the next few years.

Backup

Backups are a problem, everywhere and all the time. You had to back up the machines before, so you have to continue to backup the machines now.

Hidden benefit: Virtualization opens additional options for you to actually do the backups. For instance backing up the virtual machine images from outside the running machine or backing up the original machine image and redo logs of changes.

Provisioning

I don’t fully understand this point in the article. I think the issue here is when you have a mix of physical and virtual machines where you’re deploying applications.

Hidden benefit: if you can manage to go fully virtual, with all your applications, then provisioning becomes vastly simplified. You’ll be able to scale resources on your systems based on actual usage, rather than just sizing projections that are often fuzzy at best.

Security

The article’s point is that security is more complex in a virtualized environment and that’s certainly true.

Hidden benefit: It opens the door to a number of new options that could be very advantageous. Many of the security appliances that are currently deployed on proprietary hardware are going to migrate to being virtual appliances and that will reduce the cost and alter the equation on how it’s possible to put them to work. You’ll be able to bring the security closer to the applications and the different cost structure for virtual appliances will allow the creation of a new class of security device that would have been impractical before. You also further reduce power, cooling and rackspace requirements by getting rid of the security appliance vendors custom hardware while maintaining the ease of use that an appliance gives you.

Monitoring

Again, added complexity is the point. You now have to monitor both the base OS and the virtual machines.

Hidden benefit: Like with security, the door is open for a new type of monitoring system that can be more proactive about not only monitoring systems, but also playing a hand in recovering them. You’ll also be able to partition your monitoring needs and utilize the different information to help better diagnose where the real problem lies. For example, if one of you virtual machines loses its network, but all the other virtual machines on the same physical machine still function than you automatically can rule out a hardware failure on the network card.

Related Posts

Leave a Reply