Chris Crosby is Founder and CEO of Compass Datacenters.
Capacity planning has always been more of an art than a science; but as the annual volume of information generated, transmitted and stored is geometrically escalating, even building a facility the size of the Taj Mahal is only temporary solution. Although the breadth and speed of any expansion are critical, how to configure and connect expansion units themselves is equally important and often overlooked.
The first duty of any data center operator is to maximize uptime for providers, users of the cloud, and cloud-delivered services. This requirement is magnified significantly due to the volumes of data and the wide range of users. Unfortunately, there is no such thing as 100 percent data center reliability regardless of what an advertisement or Service Level Agreement (SLA) might promise.
A basic element of scaling up the size of a facility is the use of a structure that minimizes the negative impact of something going 'bump in the night'. We refer to it as compartmentalization, defined as the means to separate things into specific areas for specific reasons.
In terms of cloud applications, and the data centers that support them, it is the ability to physically segment those applications supporting hardware and networks in a way that eliminates the potential for proliferation in case of a service-altering incidence. Optimally, there are two elements in a compartmentalized architecture:
- Storage/compute cluster(s)
- Network nodes
Clusters are one or more groupings of like applications supporting network and hardware within the physical data center. For example, a company elects to divide the components that support a single application/business unit into distinct 1MW units or clusters. The purpose of this division would be to facilitate the ability of their technicians to identify, diagnose, and correct a problem in the shortest period of time possible.
Network nodes reside within the physical building to support one or more data halls. Each node can then support multiple clusters. These capabilities combine to minimize the negative impact of a failure. For example, in a node-supported, multi-cluster configuration, the failure of one or more clusters would be limited to only that node; thereby enabling its remaining counterparts to continue operating without disruption.
The term mesh networking is normally associated with the connectivity and redundancy between racks and the equipment they house, including: servers, storage and switches or nodes (typically top of rack). From the data center perspective, the concept of inter-meshed networking is similar. However, in this case interconnectivity resides between each cluster supported by a network node as well as between the nodes within a single or multiple buildings. For this level of inter-meshing to be implemented, each facility's network nodes require service via multiple conduit pathways, fiber access manholes and associated points of entry. Delivering fiber to each building provides the access required for support.
The primary benefit of this structure to cloud providers is the ability to isolate and contain the impact of a service-interrupting incident within a single network node without impacting other cluster-supported applications/business units within the building. This approach provides the ideal supporting infrastructure for applications that depend on maximum uptime from a standpoint of revenue and reputation. In effect, the meshed networks that link hardware and the inter-meshed structure connecting clusters, network nodes, data halls and buildings work in concert to provide the high level of reliability necessary for cloud providers and/or XaaS-delivered applications. One does not obviate the other.
The underlying requirement for inter-meshed network implementation is a structure that uses standard power specific increments or “blocks”. For example, let’s assume a 1MW block. Each block, or group of blocks in a data hall, has an independent MEP identity with its own firezone that must be serviced by multiple fiber manholes, and include multiple Points of Entry (POE). This provides access to the required fiber needed to connect clusters, nodes, data halls and buildings necessary for an inter-meshed structure.
In a sense, the continued growth of the cloud and SaaS applications, has resulted in an 'all the eggs in one basket' scenario for both large and small consumers of their services. Although any service interruption negatively impacts data center users, similar incidents in cloud environments can potentially be dramatically worse.
As a result, providers of these services are constantly seeking to implement architectures that facilitate the ability to compartmentalize their applications. This works to contain and minimize the downstream impact on customers due to any number of potential threats to reliability from cyberattacks to equipment failure. In evaluating possible service providers, it's important to understand how their architectures facilitate or inhibit their ability to implement multiple levels of containment and connectivity. Making sure network methodology aligns with compartmentalization efforts is key considering the potential injury to bottom lines and enduser satisfaction.
Opinions expressed in the article above do not necessarily reflect the opinions of Data Center Knowledge and Informa.
Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating.