Earlier this week, I attended a panel discussion that was primarily hosted by AMD but included panelists from HP, EMC, and APC and was moderated by Uptime Institute's chief analyst Bruce Taylor. The Uptime Institute earns its keep by playing host to the community of people and technologists with an interest in datacenter uptime, along with the solution providers that serve them.
The point of the panel discussion was to shed light on the widening gap between the physics and economics of Moore's Law and what if anything can be done to close it. On the physical side, chips are still advancing at the Moore's Law pace (doubling in performance every 18 months, give or take). If you spin the math, that means that the cost of some given amount of performance is halving at the same rate. The problem? The cost of the chip is now being dogged by the total cost of ownership, a part of which is the infrastructure that's needed to keep that chip running. This is primarily a discussion about energy because as the cost of energy rises, so too does the cost of running that chip. Taylor sees this as the "economic meltdown of Moore's Law."
Naturally, AMD, HP, EMC, and APC are all working on the problem and all believe to have short and long term solutions that datacenter managers should be thinking about. Other solution providers like Intel, IBM, and Sun were noticeably absent from the panel. But rest assured, they're on the case too. In the course of the panel discussion, it became clear that datacenter scale and size matter when it comes to the potential savings that could result from automating certain energy saving measures. As I listened, on question nagged at me. If that's true -- that scale matters -- then maybe going green isn't as much about buying gear from the folks sitting at the dais as it is about outsourcing your IT and maybe your entire datacenter to someone with the sort of scale that can really make a difference.
This isn't a completely senseless idea. Ten years ago, readers told us how we were on drugs when we talked about oursourcing one of a companies most precious resources (customer data) to a service provider. Today, salesforce.com -- and outfit to which such needs can be outsourced -- continues to grow by leaps and bounds. Those that chose to outsource in this fashion get to reap a variety of benefits. But in the context of the datacenter discussion at hand, one of those benefits is the elimination of the hardware and software that were in place when customer relationship management and salesforce automation we're insourced.
So, if things are going pretty swimmingly with the customer data part of the datacenter, is it not possible that at some point in the future, the rest of the datacenter will go the same path? Are we potentially looking at the death of the datacenter as we know it? Sun seem's to think so. The consolidation of the world's datacenters into four or five 'systems' is the premise behind it's Redshift Computing theory. When I last sat down for a dinner with EMC CTO Jeffrey Nick, he agreed that moving forward, EMCs had to focus on the service providers as businesses shifted their IT out of the datacenter to outfits like salesforce.com. Clearly, businesses are making the shift. And if Sun isn't right about it ending at five 'systems,' it's definitely right about the direction things are heading. More and more businesses are outsourcing to service providers which in turn can only mean one thing: fewer opportunities to sell gear. This is particularly so if you consider the physical part of Moore's Law. After all, if that wasn't the case, could you imagine the space and the HVAC that would be needed to support today's computing needs on a 1960's class Univac?
Between Moore's Law, the outsourcing of entire datacenter chunks to service providers, and the ever maturing set of technologies (eg: virtualization) designed to eliminate idle capacity, the global hardware footprint can't go anywhere but down. If you're the Uptime Institute or a gear maker like HP, EMC, or APC, this is not exactly a great trend. Even if you're Sun, it's not a great idea (see When there's only five computers left in the world, will one or more of them be Sun's?). But at least Sun, in talking about Redshift, acknowledges the reality and is seeking ways to succeed in that world. The Uptime Institutes Bruce Taylor on the other hand doesn't think the rising tide of service providers can do as good a job at running datacenters as end-users can.
In the attached video, I've excerpted the exchange between the two of us during the discussion's Q&A session where, based on a couple of salesforce.com outtages, he referred to salesforce.com as "unethical." Using the alleged unethical behavior of salesforce.com as the example, he talked about how there are classes of applications that people will never outsource: a conversation that's eerily reminiscent of the discussion that took place in the pre-SaaS days of the mid-90's. He can pick on salesforce.com for its outtages all he wants, but what of other service providers like Google, NetSuite and Amazon?
Check out the video. At the very least one of us has our head in the sand. But beauty is in the eyes of the beholder. I'm fully willing to admit that for some of you, that head in the sand appears to be mine. But I think it's Taylor's.