Virtualization for the Right Reasons

Edit: Some more good discussion. The link does not seem to work.

Originally posted December 14, 2010 on AIXchange

In a recent AIXchange blog entry, I outlined the reasons why some customers have yet to get on board with virtualization. Along those lines comes AIX blogger Waldemar Mark Duszyk, who cautions against virtualizing just for the sake of virtualization.

Here’s Duszyk:

“I do believe that there is room for a VIOS, but not in each and every data center and especially not because the admin from across the street just put one on line so we have to have it too! If you were an owner of a big and heavy track capable of heavy loads in access of 100 tons, would you use it to carry a pillow across your state? You could have used a mail service instead, right? Or if you had 10,000 of pillows to transport, you will make sure they are all compressed to fit as many as possible? The point I am making here is this you would think how to save.”

I remember watching a television show where it was argued that a diesel engine powered school bus that gets six miles to the gallon can sometimes be preferable to an economy car getting 40 miles to the gallon. You may be getting great mileage taking the children to and from school, but it takes a lot of small cars to transport as many children as the big bus can move in one trip. As Mark says, if you have to transport goods, look for the most economical way to do it. The same mindset applies to computing. Don’t virtualize just because everyone else is, do it to save on floor space and power and cooling costs, and to consolidate workloads.

 Again from Mark:

“Do I think that virtualization is a bad idea? Nope again, except that it is still a very expensive proposition. First, before even thinking about virtualization the surrounding IT environment must be comfortable with SAN boot, because without it will be very difficult if not impossible to fully utilize the processing capacity of hardware one wants to virtualize. Why? How much will it cost you to buy just one CPU (including its activation costs) + RAM + physical I/O adapters for your planned VIO environment? Now, multiply this number by two if you want to have two VIO servers in the new managed system? The point to remember is this: For VIO to save you money you have to prove that over time you will at least be able to recover the costs associated with VIO implementation. It is already obvious that if you decided to follow the VIO crowd, in order to recover the costs of virtualization, you have to pack into your managed system as many partitions as possible. Welcome to the world of SAN boot! If your partitions cannot boot from SAN you have to provide them with local disks!”

I cannot agree more. We don’t want to use physical disks and physical adapters when we virtualize. We want to boot from SAN and run many LPARs on our frames, and then we can move workloads around by running Live Partition Mobility between our frames. 

Mark also touches on workload partitions (WPARs) as well as Nigel Griffiths’ idea about running workloads and applications inside of WPARs rather than the global AIX instance:

“Use GLOBAL [instances] solely for systems management. Don’t run workloads there, and don’t create any more users than are required. Create WPARs for each workload, and create the necessary users there. Since WPARs are inherently resource efficient, you don’t give up very much by dedicating GLOBAL [instances] to management only. The overhead is certainly much less than creating a separate LPAR for each workload.”

As I’ve said: Not everyone is virtualizing, and not everyone necessarily wants to virtualize. So what are your reasons for holding back?