On the surface, hyperscale and microscale data centers seem like opposing ideas for the future of data centers. But the reality is that they are opposite sides of the same coin, or at least complementary concepts. The goal of both sets of technology is to be able to deliver data and applications when and where they are needed, efficiently and effectively.
On demand services can enhance business productivity, improve the life of the knowledge worker and enable the Internet of Things. But in order for this end-to-end process to function well, all the pieces of the puzzle need to be in place. Both hyperscale and microscale data centers are critical components to deliver information to the end user.
Let’s first look at the current concept of hyperscale data centers.
With this model, the goal is to deliver a single compute architecture that is massively scalable. Combining the use of virtualization, the software defined data center, software defined networking, and software defined storage with standardized compute, networking, and storage nodes, the goal is to be able to continually scale the IT load of the data center to meet the demands of the users.
At the moment, when we consider hyperscale data centers we think of huge cloud services like Amazon, Facebook, Google and Microsoft, which is a reasonable presumption as they provide “cloud” services that demand the type of growth and support that can be delivered by the hyperscale data center concept.
The fundamental design of the hyperscale data center needs to take account of three resources that need to scale - compute, networking, and storage. A new deployment should be able to start off small, building a baseline configuration of servers, storage, and networking that can be added to as necessary, allowing the data center to grow with demand. This means that hyperscale isn’t just for those huge infrastructures; the basic concepts can be applied to any size data center or deployment.
It also means that the cornerstones of the data center don’t have to scale simultaneously; only those aspects of the infrastructure that need to grow to meet your requirements have to be adjusted. Because the software defined data center is decoupled from the underlying infrastructure, there is no fixed relationship between the various components. uThe most important takeaway from this should be that a hyperscale data center isn’t defined by its size but by its architecture and growth capabilities. The hyperscale model has taken off significantly in the last two to three years, and is commonly believed to be limited to giant mega-scale data centers - but this may be somewhat over-simplified.
Enter the edge
While giant data center deployments get the lion’s share of attention, the biggest growth in data center development seems to be in the edge data center - smaller facilities positioned at the “edge” of the network, where end users and devices need a quick response, and don’t want the delays associated with routing to a centralized data center.
In many ways, thanks to the rise of digital services, what was once the edge has moved front and center. Smaller data centers are being built in cities that weren’t considered good candidates for tier 1 facilities. These cities have been considered prime locations for the opening of new data centers.
Although these are smaller facilities, they are prime candidates for hyperscale model development. The demand starts off small and continues to grow. Many of the providers/operators of this class of facility have begun deploying their own cloud networks for their customers, which meshes well with the density capabilities of the hyperscale model.
These highly capable edge data centers, running on the hyperscale model, are well positioned for the growth in demand for their services and for the expansion of IoT. These edge data centers will become especially important as the number of connected devices continues to rise and the availability of backhaul bandwidth grows at a slower pace.
Hyperscale changes the way data centers are designed, microscale changes the way they are delivered
So where does that leave microscale computing? As long as the cost of backhaul bandwidth remains significant there will always be a general demand for the ability to shorten the path that data has to travel to get to the ultimate consumer. There will also continue to be a need to minimize the response time between server and end user in order to provide a quality user experience. And, of course, there will always be geographic locations where the placement of a data center of any significant size and capability is unrealistic.
Two approaches to microscale
Vendors appear to be looking at the microscale data center in two ways. The first is slotting into the current model. This gives a hierarchical view of data centers and their traffic, with mega data centers at the top of the pyramid, edge data centers in the middle and micro scale data centers at the bottom. All information eventually filters up the pyramid, but only that needed by the lower tiers filters down. This approach builds on the current investment in data centers and allows for deployment if specialized, microscale data centers as needed.
The second approach looks at the microscale data center in a multiple node model, where instead of feeding back to a central facility, the individual microscale data centers are effectively in the role of an individual node in a hyperscale facility. Data and services, for example, can be stripped across the multiple small data centers, giving a level of resiliency and availability that would otherwise require specialized services and/or facilities in the current data center delivery model. This view requires buildout almost from scratch as the primary purpose for an existing data center in this model would be to eventually provide the data and applications that would migrate to the multi-node microscale data centers.
Design versus delivery
Unlike the hyperscale data center, where the concept represents a major change in the way the underlying architecture of the IT load components of a data center are designed, the microscale data center represents a change in the way the facility infrastructure is delivered.
Microscale data center designs are focused more on the packaging than the contents. Power and cooling must be delivered in a small package, often self-contained, to the IT load hardware, and this defines the design constraints of a microscale data center. However, there is no reason why the concepts that define a hyperscale facility can’t be applied to some extent in the actual compute, storage, and networking components of a microscale design.
Planning for the next generation of your data center design will require considering all aspects of information use and delivery. Building a flexible data center infrastructure that matches the flexibility required to drive business growth will likely require considering many different architectural issues in your design choices.
A version of this article appeared in the February/March issue of DCD magazine.