Posted by: Eric Siegel
After a few months of rumors, Cisco made its Big Announcement today: The Cisco Unified Computing System (UCS) is here. (See http://www.cisco.com/go/unifiedcomputing)
Based around Cisco's low-latency, lossless 10 Gb/s network fabric, it can grow to 40 chassis with up to 320 Intel Xeon compute nodes and thousands of virtual machines. It is sold as a complete system, and Cisco kept emphasizing that it would be scalable and managed as a single system.
I immediately thought of my previous blog entries (see Cisco the Computer Company, Act III) that had predicted Cisco would become "the intelligent, coordinating backplane" of a multi-server system. And I thought, gee, I was thinking too small -- Cisco is going to sell the whole smoking ball, the whole mainframe, not just the backplane.
But I kept thinking... and problems appeared. This is starting to look like less that it seemed at first. Maybe I was thinking too big, at least for the current announcement.
And the crux of the issue is systems management. Yes, Cisco talked a lot about how UCS is "managed as a single system." But the Tandem Computers (now HP Integrity NonStop) system really IS "managed as a single system." To the operator, it appears to be a single mainframe, even though it may contain hundreds of CPUs and a hypercube embedded switching fabric. I think that the Cisco UCS looks more like the old IBM SP ("scalable POWERparallel") mainframe, which was a rack with a lot of IBM AIX processors and a fast interconnect. It surely wasn't a single mainframe; the operators and users were very aware of the fact that there were a lot of separate processors there. On the Tandem, the entire thing is configured, backed up, managed, and programmed as a single machine. On the Cisco UCS (and, to a great extent, on the IBM SP), you can create templates, or service profiles, for processors, but you're still very aware of each individual processor in the system and, for example, if you want to back them up, you'll have to handle each one separately, I think.
So the Cisco UCS isn't a mainframe; it's a collection of blade servers around a high-speed interconnect with a nice template-based management package to handle configuration and provisioning. (BMC Software's BladeLogic is tightly integrated with the Cisco UCS Manager, and apparently will provide a linkage into BMC's Atrium configuration management database, etc.)
And there were no signs in this announcement that Cisco has implemented the idea of creating an "intelligent, coordinating backplane, where programmers could drop an inter-process message into that backplane (for example, with a SOAP header giving the name of the destination process) and be sure that it would be delivered to that process exactly once." Or that they've developed a diagnostics package that could easily trace transactions across that coordinating backplane.
Cisco could really shake up the computing industry if it introduced a real way of making a single mainframe image out of a collection of virtual machines and processors and network fabric. But this doesn't appear to be it. Instead, this appears to be a minor evolution of existing systems; the news is that Cisco is entering the market, not that Cisco is introducing a smoking hot new technology.
And see the Burton Group's DCS blog "Beware the Ides of March..." for more discussions of Cisco UCS!
(Note: Ten years ago, I worked for Tandem Computers as the coordinator and technical lead for all networking specialists worldwide, and I still have a warm spot in my heart for the place and for the famous Beer Busts on Friday!)
Comments