I’ve just been reading an article on capacity planning for virtual servers. I’m going to examine the issue a little deeper here and take a look at how cloud computing can help eliminate this issue.
Lately, it seems like everyone has been crowing about the number of virtual machines they manage to cram onto a single physical server. Just a few days ago, I was reading about one supposedly 100% solar powered data center who claims 70:1 (that’s 70 virtual machines to 1 physical server). Now this might look great from a power standpoint, but I’ll bet that at least 10 of those virtual machines run like garbage.
If you’re running tiny, lightly loaded virtual machines, you might get away with something like this. The problem is, most machines in the real world don’t run this way. It’s well and good to do a server load analysis, but this is a static result. Loads can and do change dramatically, frequently for totally unpredictable reasons (witness the social media overloads after Michael Jackson’s death). With a standard virtualization arrangement, this leaves you either leaving a LOT of elbow room to absorb the spikes (I’ve seen ratios as low as 6:1) or staring at the screen, manually monitoring and adjusting things. Not a very efficient or effective process, either way.
Cloud computing provides 3 easy answers to this problem. The first is load balancing. Load balancing in a computing cloud allows virtual machine workloads to automatically be shifted between the multiple physical machines of the system to avoid overload “hot spots”. No manual intervention required.
The second is virtual machine spawning. In this case, extra virtual machines are created to share the work of the highly loaded virtual machine. More lightly loaded virtual machines are much easier to balance well than a few highly loaded ones and they absorb unexpected load spikes far easier.
The third, and most important, is to leave adequate elbow room to help absorb load spikes. Even with the automated load balancing, pushing the computing cloud’s ratios up to 70:1 means that the load balancing system may not have time to react and adjust before the virtual machine overload impacts the entire host.
Of course, dealing with cloud load growth is really just as simple as adding another host to the system. Yes, it means that the number of physical servers isn’t perfectly predictable, but with the savings at even a 14:1 ratio (which we run on the SwiftWater Telecom cloud), nobody should be complaining about adding extra host server to a cloud.
Virtual server capacity planning doesn’t have to be painful or complicated at all.
Are you looking for cloud computing services, virtual servers, workstations, desktops, or file servers? Call or email me or visit the SwiftWater Telecom web site today!