Next Stop for Data Center Infrastructure: Modularity

To hear partisans of modularity tell it, the data center as we know it will be all but unrecognizable within a decade. And while I’m usually leery of claims predicting such bold changes in a relatively short time, I have to admit that this time it seems not only possible, but probable.

IT development has long been driven by twin goals: simplicity and greater power. Too often, however, the need for the latter outweighs the desire for the former, which is why many organizations struggle with bloated, cumbersome infrastructure. Modularity satisfies both needs very well, however, providing a streamlined architecture that is both scalable and easily manageable, not to mention dramatically less costly.

Already, serious venture capital is swirling around companies like Nutanix, which has devised an x86-based integrated storage/compute cluster device that promises to bring the benefits of modularity to smaller and medium-sized enterprises. As enterprises look to build new cloud architectures quickly and easily, the notion of a “data center in a box” is expected to have broad appeal in that it allows resources to be provisioned at a fraction of the time and cost of traditional data center infrastructure.

At the same time, a company called SimpliVity recently unveiled its new OmniCube, a 2U appliance that packs two six-core Xeons, 200 GB of solid-state storage and upwards of 24 TB of disk storage all tied via the PCIe interface. The purpose of the device is to federate all IT resources under VMware virtual machines through techniques like global deduplication and compression. In this way, server, storage and networking resources are “assimilated” under the device’s OmniStack architecture where they can be linked across data centers or across clouds.

This kind of modularity dovetails nicely with the new containerized approach to data center design that many of the top enterprises like Microsoft and Google have embraced. One problem, though, is the fact that network architectures within these containers are not well equipped to deal with multiple hardware failures. If only a portion of its systems go dark, the entire container could be put out of commission. However, researchers at China’s National University of Defense Technology say they have devised a new hybrid network that both simplifies network architectures and provides a more fault-tolerant environment.

But modularity is not just about providing cheaper or more robust data environments. It could also remove much of the human element that enterprises rely on to keep systems operational. As ZDNet’s Jack Clark points out, once you’ve gone modular the next step is full robotization. Imagine a data center with row upon row of modular compute devices. Automation software keeps track of data flow and performance metrics, preventing or repairing problems when it can. In the event of a true failure, specialized robots would be perfectly capable of pulling the bad module out and inserting a new one — no human hands required. A company called Kiva Systems is already doing this with standard warehouse containers, and it was recently acquired by Amazon.

Ominous stuff to be sure; although, it’s unlikely that we’ll see anything quite so drastic at least for the next two decades. In the meantime, modularity seems poised to become the next major trend in data center infrastructure, if only because it is the quickest and easiest way to put new resources in service to the cloud.