There are two trends happening in the IT hardware market, each gaining momentum but offering very different ways of outfitting data centers.
On the one hand, companies with enormous data centers such as Facebook, Rackspace, Google and Goldman Sachs are creating their own compute, storage and network devices using cheap, commodity components. The pieces are built to a standard - organized by the Open Compute Project (OCP) - to ensure they interoperate, and they are then are assembled to create hardware that is finely tuned to the specific needs of an organization. This "disaggregation" of hardware allows one company to have a system that is optimized for high-storage capacity with low CPU, for example, while another company could customize the hardware for intense reading capabilities, but low writing.
[RELATED: How Facebook aims to reinvent hardware]
Contrast that with another trend gaining momentum: convergence. Systems like the vBlock from VCE combine technology form EMC, Cisco and VMware, while other companies like Simplivity, Nutanix and others offer converging hardware components for compute, networking and storage in a single system, optimized to work together and packaged as a single offering. These "data centers in a box," as some call them, reduce the complexity of installing hardware, proponents say, and allow for easy scale-up.
"They are diverging paths, but they're both happening for the right reasons," says David Cappuccio, who advises clients on data center designs as a Gartner analyst.
The convergence trend is being fueled by legacy vendors who want to own more of the data center stack, he says. There are benefits from the customer standpoint, too, though. If an enterprise IT department is an HP, Cisco or IBM shop, getting converged hardware from their vendor of choice gives them one throat to choke, and a single vendor relationship to manage. It's a "best of brand" approach of buying hardware from a single vendor versus a "best of breed" model of getting hardware components from multiple vendors, he says.
Simplivity has created the OmniCube, a hyper-converged system that does the work of up to a dozen appliances including compute, storage, networking, deduplication, backup and WAN optimization all in one. Instead of buying multiple hardware devices from many vendors, Simplivity sells a single system that's hypervisor and hardware agnostic. "It's a stack that takes commodity resources and harnesses them together for easy management," says Doron Kempel, founder of the company.
Meanwhile, OCP backers are looking to make hardware cheaper and more standardized. Large web-scale shops work directly with original equipment manufacturers (OEMs) to get the cheapest price point, then buy in bulk. They assemble the systems using cookie-cutter-like precision. Most recently, at Interop this year, OCP announced it would begin working on open source networking components, in addition to the compute, storage and data center design work the OCP has focused on.
Rackspace COO Mark Roenigk says his company's work with OCP is all about scalability, efficiency and sustainability. The company is constructing a new data center built to OCP design standards that are customized to the specific needs of Rackspace's cloud computing and managed hosting use cases. Rackspace puts its hardware through different use cases compared to other members of the OCP like Facebook and Goldman Sachs. But, because all the components are built to a standard, each of these companies can customize the installations to their specific needs. Then, when it comes time to replace faulty hardware, it's as simple as plugging in a new compute server component, without replacing an entire new proprietary server. "We're taking the Open Compute platform, and we're modifying it to fit our specific business needs," Roenigk says.
All that OCP work is fine but it's being done by big corporations that are the giants of the industry, Cappuccio says. Most small to midsize enterprises likely don't have the resources or the desire to procure hardware components directly from OEMs. OCP organizers are trying to change that.
The goal of the OCP is not just have it be reserved for the big guys, but instead to democratize it to any size company, says Frank Frankovsky, a founding member of the OCP and Facebook's vice president of hardware design and supply chain management. The OCP is working with third-party integrators and consultants who act as an intermediary between the OEMs and the end-user companies in setting up these systems.
These do-it-yourself hardware options aren't right for everyone, says Eric Hanselman, chief analyst of 451 Research; it depends on the level of hands-on work organizations want to do with their hardware. "If you have the technical depth, it can work well," he says. Other organizations may like the idea of having an all-in-one approach from the convergence vendors.
In the meantime, as more and more resources are being outsourced to hosted environments in the cloud, this could all become moot for an increasingly large number of organizations. "The big question is, are you going to continue doing things the way they've always been done in the past, or are you going to start going down a new path?" Hanselman says. If there's a new path, then there are many options.
Read more about data center in Network World's Data Center section.