Who knew that the familiar Economies of Scale rules we all grew up on don’t always apply to the data center? While we all learned that per-unit pricing decreases as the volume increases for nearly everything consumable in life, what we didn’t take into account were the timeframes associated with consuming those quantities to make the rule valid. And in the case of a data center, these timeframes could be much longer than ever imagined.
For the data center, Economies of Scale just seemed to make sense 5 or so year ago. Simply build a data center as big as you could imagine needing for say 10 years or so, and then move into it gradually over time. Fill that huge data center with lots of high-capacity gear, and then rest easy that no matter what tomorrow brings, you are going to be all set. Lots of head-room means almost limitless capacity, on the ready, at the drop of a hat. Best of all, the world of IT rarely had a great deal of fiscal accountability, as long as they delivered “Five 9’s” of availability and uptime. So over building and over provisioning the data center was greeted as a hero’s welcome. Frankly, nobody cared if the servers were running at less than 5% utilization. No body cared if the data center was so cold you needed a warm coat just to walk through it.
In the financial justification process, all of the overview numbers just made sense. Add up all the costs to build, fill and operate the data center, then divide by the amount of TOTAL capacity to be delivered. Oops. There’s that pesky critter. TOTAL CAPACITY USAGE IS NOT GOING TO HAPPEN ON DAY 1. In some cases it is not even day 1000! So in reality, the burdened cost of the first square foot of space or the first server is in the MILLIONS of dollars, and then over time, perhaps 5 years or more, the cost of the last unit is essentially free. It is the time element that trips us up. Time to get back to the drawing board…
Enter, Stage-Left: Modular Computing. All the way from the cement up to the IT functions has now been modularized. Want modular structures? You can get modular designs that can start small and can grow in as little as 20 rack increments…. in both brick and mortar and manufactured designs. You want modular servers? Lots of flavors of those too, from nearly all of the mainstream Tier-1 players. But let’s be smart here. You have to think about modularity from a commercial and supportability standpoint, not an academic or science-fair approach. Do you really want a modular server platform, and then a modular storage platform and then modular network and security platforms? Sounds like a lot of platforms to me. Modularity is great, but some real-world life-cycle considerations are important.
Think function convergence. Along with modularity you’ll see it’s not just about scale, it’s about applying new ways of dealing with dynamic growth and expansion. It’s about converging multiple functions into fewer and more easily supportable physical platforms that have higher utilization and then deeply federating these converged offerings to make the aggregation of capacity supplied from many building blocks indistinguishable from those that would be supplied by a single BIG unit when viewed from the outside. Think of placing two smaller drops of mercury on a table top, and pushing them together with your finger (OSHA warning: don’t try this at home). When the two drops of mercury touch, they instantly and completely bond to create ONE drop of mercury. Take 100 drops and push them together, and you’ll still end up with ONE really big (and shiny) drop of mercury. In the case of mercury, its their molecular bonds that allow this. In the case of IT, it requires some pretty innovative engineering to create the equivalent of molecular bonds across devices.
That is what modular computing is all about. Forget most of the rules associated with economies of scale, or at least consider these economies of scale within a finite period of time. Everything else becomes a modularity discussion in the context of time.
Whether you are thinking about the structure itself, or the servers, storage and network that are contained inside, you really CAN think modular today. There are companies that offer their wares which act like drops of mercury. With these offerings, you can add a few thousand square feet of space to a properly designed data center with the result behaving as if it were a single center. I can deploy a handful of converged IT building blocks that each combine their needed networking with processing and storage and do so over and over again, each time federating these devices into what appears to be a single structure.
With patented technologies and intentional design features, modularity is not only possible, BUT VERY COST EFFECTIVE when viewed over the same period of time as any data center is intended to be in service.
(References: IO Data Centers, Compass Data Centers, Simplivity, VMware)