The Recent Cisco UCS Announcements (September 4, 2014)

Peter Welcher
Architect, Operations Technical Advisor

It was great to have the chance to attend the September 4, 2014 UCS GrandSlam event / launch with Tech Field Day. One indicator of the importance of the announcements was the copious e-press coverage before and after. There were several announcements at the launch, so I’m planning to discuss the launch one component at a time, work priorities permitting, in a series of blogs.

One thing about this launch that impressed me was the relatively low amount of high-level jargon. Some launches are just prime material for buzzword bingo (and don’t forget that “paradigm” is a double wildcard). I’m never quite sure if “ecosystem” qualifies as a buzzword — Cisco uses it a lot, but having a strong and big ecosystem is clearly a market differentiator for Cisco as well.

What Did Cisco Announce?

The UCS M4 series servers, 4th generation UCS. More powerful, smaller. Perhaps not a big surprise, more of an incremental update, but well done.

  • B200 M4 blade server: performance and density
  • C200 M4 rack server: general purpose and density
  • C240 M4 rack server: storage and I/O for big data and virtualization
  • VIC 1300: next gen convergence and virtualization

UCS Director manages all the announced products. It now does application containers, has Hadoop integration for large clusters, does ACI (Application Centric Infrastructure) configuration, and has an open developer kit.

The UCS Mini supports “edge-scale computing”, for small environments. Aside from SMB (where the Small-Medium references large values of “small”), the use case is retail or branch computing, or distributed server resources for e.g. aggregation of IoE data/ fog computing. Centrally managed across hundreds of locations, with simplified systems management, with low power and cooling footprint. Supports 1-15 servers, compute, networking, UCS manager, in one 6 RU package, with expansion by up to 7 C-series servers. The UCS Mini has been shipping for several weeks. And there’s been some previous discussion around the small 6324 FI Fabric Interconnects.

I’m repeating some of the Cisco prose here because I get this use case. I’m seeing two contrary trends. Some customers are trying to get servers consolidated and virtualized in “real datacenters”. And that can simplify operations, but requires a robust WAN. Other enterprises are going the other way, putting servers in the field to reduce latency, speed responsiveness, or avoid sending large data across relatively slow links. If your remote offices are in big cities, bandwidth is likely not a problem. But if your office or plant or warehouse is in the middle of farm country, chances are fiber isn’t available, and a T3 may be costly. Still. Unless you’ve got a civic-minded firm in the area, like my friends at NetgainIS, which built a “ring around the county”. But I digress…

The UCS M-Series Modular Servers (or as someone said at the launch, the “MMM” servers?). Personally, my thoughts jumped to “Mars Shot” (due to some similarity to HP Moonshot — but going a lot further).

The story on the M-Series is that the VIC 1300 lets Cisco decouple or “dis-aggregate” the server from storage, to provide cloud-scale capabilities. (More about that in another blog.) The marketing name for this is “System Link”, extending the UCS FI inside the server. The key point is to separate powerful CPUs and memory from shared storage and networking resources. This is useful if you want high server density, as takes place when web-scaling an application. With virtualization, the game is to pack many VM’s into one powerful CPU to save VMware licensing costs, power, space, etc. With web-scale, it’s often one app per CPU, with scale-out via adding physical clone servers. (I hope when you read “clone” you think “UCS profile” there.)

The Modular UCS uses compute cartridges, 8 per compact chassis. Each cartridge has two separate Intel Xeon E3 servers, but no NIC adapters and no hard disks. Instead, the servers can share four SSDs and 40 Gb of networking. That means you can fit 16 Intel Xeon compute nodes into a two (2) RU chassis.  That allows for a much smaller physical footprint,  a ton less cables and complexity, and removes failure-prone hard disk drives from being tied directly to the server. It also saves power. It’s all managed by The UCS Director Who Must Be Obeyed. (That was my mandatory obscure literary reference for this week.)

The last big item was the Cisco UCS C3160 Rack Server, which is a standalone server optimized for storage (complementing the modular UCS servers).

There was also mention of UCS Director Express for Big Data, single-touch deployment and unified management for Hadoop. It’s a cluster profile combining Cisco UCS service profiles with Hadoop profiles.

Related Links

People have been talking! (Well, writing…). There’s been a lot of prose about the UCS announcements. I’ve listed just some of the links I’ve seen below. (And thanks to #CiscoChampion for providing some of the following links as well.)

Somewhat Related Links

There was also a recent announcement of Cisco partnering with Simplivity. There’s been some good speculation about that, great topic for another blog. I might just avail myself of the rare chance to spread scurrilous and unfounded rumors, some more unfounded than others.

Links Which Have Nothing to Do with UCS but I thought I’d include them anyway

Just kidding.

Hashtags: #UCSGrandSlam #CiscoChampion #TFD #netcraftsmen #CiscoUCS

Twitter: @pjwelcher

Disclosure Statement

 ccie_15years_med   CiscoChampion200PX


Leave a Reply