Bare Metal Clouds and Dedicated Data Centers

Chris Crosby, CEO of Compass Datacenters

When any person, product or technology reaches the level of public awareness where nuance has given way to ubiquity (and the singularity denoted by the word “the” has become the accepted prefix to a name or descriptor), opportunity for competitors often begins to arise from their long shadows. For a number of customers, the shared nature of public clouds is an acceptable negative element of the solution’s experience, but—as in the case of dedicated data centers—the unique requirements of a sizeable segment of the marketplace have led them to a perspective that is totally at odds with your mother’s admonishments regarding the relationship between your toys and your siblings and friends, in other words:

It’s not so nice to share.

The alternative to the egalitarianism of the shared cloud is, of course, the vaguely Teutonic-sounding “Bare Metal Cloud (BMC).” Does anyone think this would look better with umlauts? Spawned from the fertilizer of dedicated hosting, the original SoftLayer crew coined the term back in 2007. Does anything better say, “hands off, this is mine”, than the conjured imagery of your servers housed within the walls of a fortress of cold, hard steel? Of course not, so I think the line of market demarcation is pretty clear in this case. Really cool naming conventions aside, the emergence of the BMC market parallels that of dedicated data centers in that, for many companies, the drawbacks of shared environments provide unacceptable limitations on their ability to administer their mission critical applications.

In the normal world of the cloud virtualization, multiple workloads may be virtualized on a shared physical server. Although this is an acceptable solution in many instances, when multiple applications are all in contention for the same processor and memory resources, the risk that these same resources might not be available when they are required by a latency sensitive application (the latest Call of Duty for example) can be intolerable. The performance issues within a shared cloud may also be exacerbated by the processing capacity consumed by the hypervisor layer that can consume as much as 10% of the available capacity of a server. In a sense, this sensitivity is analogous to the reasoning behind many organizations’ desire for a dedicated data center that they can locate as close to their user community as possible (edge facilities, for example) to eliminate the latency delays associated with using more centrally located shared mega facilities.

In the case of prospective BMC and dedicated data center customers, the access to the computing platform is of paramount importance to the ability to support mission critical applications. The inability to control the risk of resource contention in a shared cloud environment that is incompatible with the needs of the BMC customer also approximates the security concerns that typically compel businesses to pursue a dedicated data center solution. In both instances, the risks associated with operating in a shared environment are incompatible with the mission critical requirements of the prospective end user thereby making the enhanced security and control provided by BMC, or a dedicated facility, the only acceptable alternative.

In a shared data center, end users knowingly sacrifice control over elements such as the site’s security system, IT or network as these elements are of lower importance in operating their computing environment. Dedicated data centers provide an alternative to these sites for customers whose operations dictate that they have total control over the operation of the facility. These same limitations of shared clouds are addressed by their BMC alternatives. In these instances, end users control all the risks of resource contention by making decisions regarding assignment of VM’s across physical servers, monitoring and balancing loads as needed. This higher level of oversight and configuration capability is not available within shared cloud offerings as only the provider has visibility across the entire, multi-tenant environment.

Specific end user requirements require the ability to “personalize” their solutions to meet their unique needs. In a shared environment, be it a cloud or data center, deviations from standard offerings are often difficult and expensive or completely unavailable. In the case of BMC’s, or dedicated data centers, the ability to “personalize” the final solution specs is imperative to support their end user’s mission critical application(s). In BMC’s, the process begins with the physical server serving as the basic unit that customers are able personalize by specifying items like the processor type, memory and storage. This capability is replicated in a dedicated data center through the ability of end users to select from multiple potential capabilities including the ability to mix and match data hall size and power options. In both instances, the end user is then able to more efficiently deploy and service the servers under their control.

In many ways, the processing requirements that determine the need for BMC or a dedicated data center are very similar. In both instances, high volume processing is a paramount requirement. For each, the data being processed in mission critical in nature and is often associated with the manipulation of the large volumes of data or high performance capability. The changing nature of the types of data that move through both, including the billions of tiny packets associated with the IoT and large, rich packets that characterize applications like streaming video, has made the elimination of latency a prime motivator in the decision to shun shared alternatives. In the case of BMC, the driving factors are primarily capacity and access related while geographic location is often the solution to the problem from a dedicated data center perspective. Both cases are indicative of the rapid changes in the structure of data that are being driven by the demands of a more connected and sophisticated base of end users.

Naturally, economic considerations are also play a large role in determining the need for BMC or a dedicated facility. Just like with a shared colocation facility, the shared cloud model is predicated upon the fact that users will not use the resources effectively and thus more power (or compute) can be ‘sold’ than what is actually used. The challenge is that performance does matter, and while that usage may be true most of the time, it is NOT guaranteed to be true all of the time. Risk profiles of certain organizations do NOT dictate the need for shared infrastructure if the economic benefit is NOT significant. Just like a user with a fixed profile of usage can get just as good, if not better, cost with dedicated BMC, a sophisticated IT user with scale can get as good, if not better, cost with a dedicated data center.

An example of this principle may be found when we consider what is referred to as the “Hypervisor Tax” associated with shared cloud environments. As previously discussed, the hypervisor layer can consume up to 10% of capacity. For a low performance application, the cost of the “tax” imposed by a 10% capacity reduction may be minimal since the negative impact to end users due to the unavailability of a computing resource is deemed to be negligible. In this instance a shared cloud environment is the most effective alternative. For a company supporting a high performance workload application (a financial trading organization, for example) the cost of unavailability, in both real dollars and customer satisfaction, provides all justification necessary for selecting a BMC alternative.

The need for both BMC offerings and dedicated data centers has arisen from the changing nature of data processing and storage. In a sense, both have emerged to address the shortcomings of shared alternatives and reflect the growing need on the part of many end users for greater control and configurability in their mission critical operating environments. While it is doubtful that either will come to define the majority of their respective marketplaces, they do illustrate that for a significant number of sophisticated IT organizations, “selfishness” is a completely acceptable alternative to sharing.

Chris Crosby is the CEO of Compass Datacenters.