Virtualization Solutions – Same Motivations
I’ve been writing a lot about virtulalization lately, and the process has been clarifying. I figured out a few things, such as the fact that none of the stuff we are talking about is new. Most of it is really old actually.
Perhaps the most interesting conclusion I’ve come too is that all virtualization (computer, anyhow) solutions we’ve ever come up with have followed identical financial paths. First, they attempt to solve capital expense related problems. Second, the attempt to solve operating expense related problems. And third, somewhat related to second, they attempt to keep people from jumping off buildings – or solving people scale problems.
A storage example would be Volume Manager. VM came about because of capital economics – i.e. when Seagate came out with a 4GB disk drive that was only 1.5X the cost of a 2GB disk drive, people bought them – lowering their capital cost per MB. The problem was that the operating system could only deal with a 2GB disk, so in reality what you added was a more expensive 2GB disk – until VM made it look like (2) 2GB disks. Phase two of storage virtualization was when we came out with RAID arrays – taking a ton of disks and putting them into a single box making it look like whatever we wanted. By doing that we got huge operating efficiencies, because managing a million individual disks is harder than managing one big box. Now we are at stage three, where we have a million big boxes, and people are going to jump of the building.
We are focused on keeping people from killing themselves, so most of the solutions to the Multi-box management problem are just moving the virtualization element to the next step in the process – looking at the boxes instead of the disks, etc. That will help, but it’s not the end game, it just is like all the others, a band-aid.
Servers are now doing what storage did years ago. Servers cost too much and we have too many, so use virtual machines instead. Smart. Not new, but smart. Mainframes and big Unix boxes have been doing this since the dawn of time. Now we can do it on Windows and you’d think we’ve seen the second coming.
What we need is the 4Th motivation -which no one has figured out yet – and that is the ability to virtualize AND control the entire layer – be it storage, server, network, or whatever. That means much more than making one thing look like something else – it requires deep, intimate knowledge of all the underlying infrastructure so that while it masks the complexity from you and I, behind the scenes it controls it – the way an operating system does in your PC. Then you have real business value – that’s when we stop doing tactical IT infrastructure stuff and all our efforts go into doing something with the data we have to drive down costs, make more money, or solve world peace – something a tad more meaningful than mastering an array GUI.
This is hard, of course, as recently pointed out in this blog by Greg Nawrocki who comments about another claim about virtualization and cold fusion, by upstart Crosswalk. I read the Crosswalk release, and heck it sounds awesome, but even if you nailed one layer, you have to be able to have understanding deep enough to coordinate between all the primary virtualization layers – be they server, storage, or network.
Howard Smiths blog, Forked Path talks about all sorts of stuff that shows the difficulty of the big picture interdepencies.
This is a good virtualization specific site that spends more time on server side stuff, but it’s stuff you’ll need to know.
A big part of the problem ultimately is that the different virtualization layers have much different players, and they are solving much different problems without much concern about the other. If you build a hammer, everything looks like a nail, right? 15 years ago you bought your system, application, network, and storage from the same guy, but no one does that anymore. They were forced to deal with each other, or perish. The virtualization guys – who’s market is still too nascent, don’t have to deal with each other yet. VMware is about a close as it comes since they are owned by EMC – but if you talk to an old time VMwarian, they don’t know jack about storage, and certainly nothing about networks.
It’s really hard to make things really easy.