I’ve been catching up on some back issues of the IBM Open Systems magazine, in the latest issue, August ’07, Ken Milberg provides a useful overview of the new AIX Workload Partitions and comes to the same conclusion I did, “I see the WPAR as a real complement to LPARs, not a replacement”.
Over on Julien Gabel “blog’o thnet”, he does a liberal comparison of the new AIX and Power 6 features with some existing and many upcoming features promised for Solaris. It’s an interesting comparison insomuch that there has been much discussion over the similarity between AIX Workload Partitions and Solaris Containers. One of the reasons we introduced containers now was the linkage and exploitation of Live Application Mobility. Julien draws the distinction between Solaris Containers as utilization feature, and not a virtualization feature.
For my part I don’t see the difference, and the more you think about this, the more obvious it becomes.
In order to have effective utilization you need virtualization, and one of the best exploiters of virtualization, you need mobility. Otherwise you are left with a workload that is stuck on a single physical or logical server. This gives more grist for the mill to those how incorrectly assert that it makes no sense to run a large system, in case it goes down. First, if this is a concern you’d never go with a single server, in all cases I’ve ever heard the same potential failures would also apply to any number of rack-based servers.
In 1998, I was lead architect for an Internet banking system that used just two physical servers and four logical/virtual servers with a shared, clustered database. The system scaled from its’ original 3,300 users to a staggering 900,00+ users over 2-years without ever losing or taking both systems down at the same time. The one thing we were missing was mobility. Even though each physical server was backed up by the other, and even though each logical server was backed up, what it meant was that when a logical or physical system was taken down, there was a noticeable time delay while the OS and Middleware was started or restarted.
So, in an increasingly eco-aware IT environment, the ability to drive up utilization up, power consumption and heat production down, you need mobility. Without it, all you’ve got is a two leg stool.