I was recently forwarded an article by a co-worker entitled “Virtualization doesn’t solve any problems”. First let me say that I think the title should maybe be revised to “Virtualization doesn’t solve all problems”, but did anyone ever claim that it did? To say virtualization doesn’t solve any problems is a bit arrogant and I think uninformed. The intelligent virtualization user / implementer is aware of the current limitations such as I/O bandwidth constraints. Virtualization is not a one size fits all but it is a size today that fits a much larger market segment than SMP cluster ting. While there is a segment of the market such as life sciences that is dealing with the need for massive parallel processing and incredible I/O requirements a much larger market segment is struggling with aging infrastructure, massive under utilization, server sprawl, growing environmental costs (hvac, energy, floor space, etch..), and the need to simplify recovery. While SMP clusters are near and dear to my heart (I worked in the life sciences field tweaking code to streamline molecular modeling computational operations, we used SGI at the time Linux HPC and products like Beowulf were just not there) they still have a way to go before they become mainstream, one of the great things about virtualization is the plug-and-play befits, the ability to move from the physical to the virtual and realize benefit immediately. Often in the SMP world applications need to be modified to take advantage of the new compute power. Virtualization players VMware, XenSource, VirtualIron and others are working on hypervisors that hold the promise of the simplicity and functionality that more complex SMP clustering applications provide.
On a final note there is another interesting quote “Virtualization addresses the same problems, such as server glut and management complexity, as clustering does — and it doesn’t necessarily do a better job of solving them” I don’t think that I agree with this statement. There are soft benefits that are just as important as the hard benefits such as server consolidation, environmental cost savings, etc… VMware emerged as the early player in the mainstream visualization market as an offshoot of a Stanford project called the Hive OS, today the Hive OS project is part of a larger project at Stanford called FLASH (FLexible Architecture for Shared memory). I recommend reading about the Hive OS to understand the befits such as isolation that virtualization provides.
Anyway I think I am done for now. I would enjoy hearing your thoughts and/or comments.