With thanks to Mark Thiele (EVP Data Center Technologies) and Jason Mendenhal (EVP Cloud), I was able to spend some quality time in-and-around the facility (even inside one of the AC units) and got an education that mega data centers are much more than a structure to house servers.
First - let me give you some amazing first-impressions: Switch's SuperNAP is all about design and function. The structure, the architecture, the even the color scheme is all for a purpose - to communicate attention to detail. And while it might cost a bit more to color-code pipes and conduits, label every piece of equipment, architect custom lighting, or build with industrial design principles, it's clear that every item in the entire structure is there (and highlighted) for an intentional purpose.
And now for some observations and learnings:
In all of the talk about energy efficiency, the SuperNAP averages a PUE around 1.24 during the year. That means overall Switch's ability to minimize energy loss and to maximize cooling efficiency is extraordinary compared to most competitors... less than 25% of the total facility power is used for "overhead" operations, while the majority goes directly to the servers and equipment. That's a relatively rare feat these days, with the industry average of 2.9, and only 20% of data centers scoring better than 2.0 according to a Digital Realty Trust survey reported by the Data Center Journal.
Arguably the breakthrough for the design of the datacenter is Rob Roy's breakthrough thinking about cooling. Traditional data centers take a "diffusive" approach to cooling equipment. That is, they place AC units throughout the data center, diffuse cool air (via raised floor) over everything, and allow the hot air from the servers to re-mix with the cooler ambient air.
But if you think about it, server racks should really be viewed as radiators (think about the radiator in your car). To achieve the best heat transfer efficiency, pull cool air directly through the radiator, and channel it directly back to the cooling device. And that's what switch does with their T-scif (Thermal separate compartment in Facility). Cool ambient air is pulled through the server racks and channeled directly into a hot-air plenum. there is no mixing of the hot air w/cool. Think of this as hot-aisle/cold-aisle containment taken to the extreme.
Data Center Density
The notion of density makes sense along many dimensions. First, more server manufacturers are designing equipment as inherently dense - a packed Cisco UCS or Dell Blade enclosure can potentially pull 7kW or more. And there may be 4 or more of these units in a rack enclosure.... that means a given rack might consume as much as 15-25kw. (The SuperNAP is designed to support about 1.5kW/square foot, easily enough to handle a packed cabinet). But the beauty of density also helps with creating a larger heat differential aiding in heat transfer efficiency.
Supporting power and cooling density ultimately helps Switch because it means that a given customers needs less space - which equates to $ savings. Yet density also helps Switch due to the obvious efficiencies.
Very early in the tour Jason pointed out the advantages of customers and partners co-locating their equipment within the SuperNAP. With the physics of bandwidth and latency hard-at-work, it became clear that certain customers had an advantage to co-locate their private cloud infrastructure in the same data center as their public cloud partners (or other service provider customers). Apparently there were many examples of this co-located hybrid cloud approach at the SuperNAP.
The SuperNAP is literally a nexus of multiple Network Access Points - that's the NAP part - originally built by Enron. (In fact, it was pretty cool literally seeing the conduits come up out of the floor with the fibers inside!) This fact provides users with advantages such as network redundancy as well as the ability for Switch to broker bandwidth at wholesale prices. They market this as the Combined Ordering Retail Ecosystem (CORE).
Physical Security and Disaster Preparedness
It wouldn't be complete if I didn't mention the physical security that Switch provides... something slightly out of "24". While it's not appropriate to share details, suffice it to say that entry onto the campus, into the building, and around/within the cages was closely and carefully monitored and guarded. The facility itself is secured in multiple zones, as-are the tenant areas.
But the other security blanket Switch offers is back-up power and cooling. Should there be an outage from the local utility, cooling is literally designed to "flywheel" for enough time for generators to kick-in. And there is sufficient fuel on-site (and sources off-site) to maintain operations through even the fiercest unforseen disaster.
Example: Content Hosting and Distribution
Without naming names, a major content streaming company chose Switch for nearly all of the reasons above. But what I found particularly fascinating and compelling is that they even co-located their physical broadcast operations at the SuperNAP. Picture aisles of million-dollar storage arrays pumping-out movies and live TV shows 24x7 across the country. Co-located is the nerve-center -- not unlike NASA mission control -- that monitors performance and delivery country-wide. Why Locate at Switch? Besides all of the efficiency and security aspects, access to nationwide backbone networks ensures maximum video performance.
Going forward, Switch will be more than doubling its capacity in the next year or so, building into new expanded facilities. And from the sound of it, much of the space is already spoken for.
Who says the data center isn't important? I certainly wasn't underwhelmed by this visit.
For More Information
- Assessing Data Center Metrics Fountainhead blog
- High vs. Low Density Data Centers, Cloud, Cost, PUE & Sustainability by Mark Thiele of Switch
- Mapping a Course to Data Center Efficiency by Jack Pouchet on Data Center Knowledge Blog