In-Depth

UCS Marks Third Anniversary

Cisco calls Unified Compute System 3.0 "third generation of fabric computing."

• By Stephen Swoyer
• 03/30/2012

Industry watchers say Cisco and UCS have accomplished quite a bit in those three years, including a promotion -- Gartner Inc. recently moved Cisco from "Visionary" to "Leaders" in its "Magic Quadrant for Blade Servers." Cisco also enjoyed strong revenue growth, with 2011 UCS-related revenues of $1.3 billion and its 11,000th UCS customer.  All point to UCS' uptake, momentum and impact, Cisco says.

Cisco does have important milestones to brag about.  Industry veteran Dan Kusnetzky, a principal with consultancy Kusnetzky Group, writes that "Cisco does appear to be growing strongly and has moved its position in the worldwide server market from zero to over $1 billion in revenues in about three years -- healthy growth by any measure."

Kusnetzky and other observers note that the recent UCS 3.0 launch's emphasis on fabric computing also puts the spotlight on challenges facing the company.  Although Cisco has been pomoting fabric since 2009, it's not the only vendor doing so.  It must be mindful of data center infrastructure competitors. That's one reason Kusnetzky is skeptical of some of Cisco's claims about innovation. As part of its UCS 3.0 release, for example, Cisco cited both technology integration (unified compute, networking, and virtualization, storage) and investment protection (including an easy upgrade path to new UCS resources) as competitive differentiators. The problem, Kusnetzky argues, is that several competitors (including Dell Inc., Egenera Inc., Hewlett-Packard Co. (HP), IBM Corp., NEC, and Fujitsu) can make similar claims.

Rivals IBM and HP surpass Cisco when it comes to management features.  "IBM and HP both offer extensive portfolios of management software, allowing their systems to appear as pools of dynamically assignable resources," he points out. "Each of these suppliers offer systems that treat system resources as a pool that can be assigned to tasks as needed and offer sophisticated management tools to make the process simple," Kusnetzky points out..

In the data center networking market, Cisco has made fabric a centerpiece of its recent product development. Here, too, it faces plenty of competition. Rivals including Arista Networks, Egenera, HP, IBM, and Juniper Networks all say they deliver innovative fabric technologies.

At the high-end, Cisco has had plenty of competition from arch-rival Juniper, which announced its next-generation data center fabric technology, QFabric, in February 2011. QFabric wasn't released until Q3 2011, and Juniper was able to tout a signal customer win -- Bell Canada -- by late October. At about the same time, Cisco released a new Fabric-2 module for its Nexus 7000 switching line. Fabric-2 offers about double the performance per port of its predecessor (Fabric-1) and uses a parallel forwarding technology to boost the overall capacity of Nexus 7000 to 15 Tbps.

QFabric looks powerful, at least on paper. Although Juniper claims a scalability of over 60Tbps for Qfabric, benchmark results the company released last month show a maximum of roughly one-quarter of this speed, or 15.3 Tbps for a total of 1,536 10GigE ports. That's nearly identical to the maximum capacity that Cisco claims to deliver with Nexus 7000 and Fabric-2.

The recent UCS 3.0 launch also highlighted how Cisco is leveraging its fabric technology to scale its UCS server assets. Cisco announced its new UCS 6296UP Fabric Interconnect, which offers double the capacity of its existing fabric (1.92 Tbps versus 960 Gbps) and which the company says can be used to scale UCS deployments by connecting Nexus Fabric Extenders (FEX). Enterprises can tap FEX to connect up to 20 UCS chassis in a single system.

The Price Is Right

Dale Roberts, a manager of platform support services with a prominent university based in Washington, D.C., says he wasn't attracted to the potential management synergies of a combined UCS data center and Cisco networking stack. (Roberts' employer is a big Cisco networking shop.) Roberts wasn't overly concerned about the switching capacity or performance of Cisco's data center fabric. As Roberts and his employer saw it, the price had to be right.

At first, it wasn't. "Entry costs to UCS are [prohibitive]," he explains, noting that this is chiefly a function of the cost of UCS chassis itself. "That's why we ignored [Cisco] before. Then they gave us the chassis for free, which reduces the cost dramatically," he continues, adding that the UCS "blades themselves are pretty reasonably priced." Roberts' shop transitioned from a SPARC-based infrastructure of standalone or rack-mounted systems to a UCS cluster running VMWare and Red Hat Enterprise Linux (RHEL). Although cost was the primary driver, Roberts says he's come to appreciate the built-in upgradeability of the UCS paradigm.

"[W]e will never have to buy fiber [i.e., fabric] again, [because] we connect it once and [we're] done," he concludes. "[Basically,] they gave us a sweet deal, and [UCS has] got a lot less cabling than a cluster of standalone servers and ... other blade servers."