Cisco is taking its shot at data-center hardware disaggregation with the M-Series, being launched today as part of a major announcement for the company’s Unified Computing System (UCS).
The overall theme is that UCS is expanding its scope. That means it’s reaching outside the data center (see the UCS Mini description below), but more ambitiously, it’s about prepping UCS for what Cisco hopes will be an onslaught of big-data applications. As you might guess, it all ties back to Cisco’s Internet of Everything (IoE) obsession, where all manner of devices generate huge volumes of data to sift through.
Joining the Disaggregation Club
The M-Series is interesting by itself, IoE or no IoE, because it’s a new form of server architecture. An M-series system is a box 2 rack units tall with eight slots for cartridges of Xeon processors (with accompanying memory) or hard disk drives. There’s also some solid-state memory shared by all eight slots.
What you’ve got, in a sense, is a blank box that can be filled with computing and storage that can be applied to applications in varying quantities (although the combinations are limited; only four of the eight slots accommodate disk drives).
It’s similiar to Intel’s Rack Scale Architecture, which piles compute and storage nodes into a rack with the intention of accessing them as pools of resources. A key difference, though, is that Intel is working at rack-scale, whereas Cisco’s M-Series is more about rethinking the innards of a server.
“We’re not going after the hyperscale space here,” says Todd Brannon, managing director for UCS. “This product is not built for Facebook or Google or an Amazon where you have hundreds of thousands, if not millions, of systems. This is for that next 500 or 1,000 customers down that scale into the thousands or tens of thousands of servers.”
The driving principle is that even for that next tier of users, applications no longer fit nicely in one server. In the era of scale-out cloud computing, an application can be spread across many nodes.
It’s also not Cisco’s answer to the Open Compute Project, as the latter is more of a “physical packaging exercise,” Brannon says. The M-Series, on the other hand, is about letting the “size” of a server stretch or shrink to fit the needs of an application.
Cisco UCS Spreads Out
The rest of today’s announcement represents a territory expansion for UCS. Cisco describes it more as a way to present a uniform operating model, both throughout the network (data center to the service-provider edge) and throughout the portfolio of apps being used (from office stalwarts to newer, scale-out applications).
Here are the other major pieces being announced:
- A fourth generation of UCS servers: new two-socket machines in rack and blade form.
- New capabilities in the UCS Director management platform to accommodate the scale of big data.
- UCS Mini, a platform for the edge network outside the data center.
UCS Mini is worth a few more words. UCS is designed to be a fabric; UCS Mini shrinks that fabric interconnect into an all-in-one system. It’s meant to sit at a remote source of demand, processing data without having to transport it back to a data center. An oil rig would be one example, but since that’s becoming the industry’s overused, go-to use case, let’s try this one: A connected city could use a UCS Mini to process citywide sensor data locally before sending it to the cloud.