Issue link: https://cp.revolio.com/i/661564
THE DATA CENTER JOURNAL | 3 www.datacenterjournal.com There is a popular myth that purports cloud computing is making data centers irrelevant. That's simply not true, but it is safe to say that cloud computing is making data centers dramatically different. To me, it really boils down to two key changes: density handling and adaptability. F irst, cloud computing is very dense in terms of power requirements and heat production per square foot. Around 90 percent of data centers today are incapable of handling the heat density generated by cloud computing because they weren't designed for it. As my chief data center designer Todd Gale says, "Any idiot can put a lot of power into a small area; the trick is getting the resulting heat out." In older data centers, the solution is typically to spread those high-density cabinets out because you can't cool that much power in such a small area. But with the cloud, you don't want to spread out those racks because doing so creates more cabling costs, including optics and intercon- nects, as well as increased latency. Low latency is especially critical in converged and hyperconverged cloud technologies, limiting the ability to spread out cabinets for cooling reasons. A final drawback to spreading out the cabinets is the resultant cable sprawl, which introduces manage- ment complexity, making it easier to introduce inadvertent downtime when performing maintenance or upgrades. Some modern data centers—what I call Generation 3 and 4 data centers— have high watts-per-square-foot cooling capability. at cooling is necessary for highly virtualized environments, which can oen generate 15k, 20k or 25k watts per rack. Yet most data centers that were built even just a few years ago (and unfortunately some being built today) can't handle that much heat. e reality is that if you haven't built a data center to handle this kind of heat density, then restructuring it is seldom viable. It really needs to be built in from the ground up. Second, cloud computing is vari- able in terms of load and location. You may see fluctuations where it is very demanding on resources during some periods, but there are also periods where demand shrinks or moves to another location. For data centers to be economi- cal, they must be able to adapt better to change. e data center needs space, power and cooling available as cloud pods need more resources, but it also needs to be constructed so it can redirect those resources to other customers or uses when those pods shrink or move. One approach is to think in terms of large homogeneous pools of space, power and cooling. Rather than creating high-, medium- and low-density areas, make all areas capable of dealing with high density, but also capable economi- cally of being low density to accom- modate cloud computing's variability. Previous approaches, where space is tied to power and cooling in smaller chunks (think 10,000 square feet tied directly to 1.1MW), required that the designer guess the ultimate power density of each space, tying more power and cooling to high-density space or less power and cooling to low-density space. Inevitably that guess is wrong. Usually the guess is wrong initially, but it's almost surely wrong over time, stranding either space or power/cooling. Large, homogenous pools of resources, where space is not tied to power and cooling, allow the law of large numbers to help create much more efficient use of resources, reduc- ing the stranding of those resources (and stranding of the capital required to deploy those resources). Both of these needs—density handling and adaptability—require a significant change to the way data centers have been designed in the recent past. e process requires facilities to be rebuilt or redesigned from the ground up. Companies hosting their cloud in data centers, even those that are just a few years old, should be cognizant of these aspects, since it will be difficult for existing data centers to change in order to meet these two goals. is need for newly-designed and built data centers has led some to believe that all cloud computing is going to end up in the data centers of the hyperscale behemoths, like Amazon, Google and Microso. But many others believe that a solid portion of cloud comput- ing will continue to happen outside of those big companies—whether through individual enterprise cloud computing or nonpublic shared cloud computing. Although the big players certainly have large footprints, they don't provide all of the functions and services that small and medium-size businesses require, and they don't provide anything at all to those companies requiring a hybrid com- puting solution. is situation creates an oppor- tunity for other cloud providers that can address those additional needs. e underlying key for these other providers, though, is building data centers that can handle the density and flexibility require- ments that today's cloud computing demands in order to future proof in an economical manner. n