Update on Cisco’s UCS

by Robin Harris on Thursday, 2 April, 2009

Brief update on the Cisco WebEx conference this morning. I got into the conference late and missed the embargo requirements, so this could be above your pay grade.

Also, WebEx wasn’t showing me the slides until late in the presentation. But here’s what I gleaned despite the technology.

A key piece of the story is something called VMlink, which can see and manage VMs throught the network. It can also attach policies to specific VMs that follow the VM as you move it around on the infrastructure.

Large memory support
Cisco has added the large memory support through some unspecified technology – but I bet StorageMojo readers have some ideas. The added DIMMs are industry-standard, so there’s no Cisco lock-in associated with them.

Up to 40 some-odd DIMMs can be added. They don’t expect that people will use the costly 8 GB DIMMs to max out memory, but rather use lower-cost DIMMS to get beyond what standard Xeons offer at a lower cost per GB.

The system will support up to 320 blades in multiple racks. This goes well beyond the 32 VMFS supports in physical servers, but Cisco expects few customers will get anywhere near that size installation before that number expands. A safe bet considering how long enterprise data centers test before deployment.

Their market focus is on applications requiring many VMs and large data sets.

The StorageMojo take
VMlink sounds good, but like my WebEx connection, a lot depends on the implementation. Large memory support is a Cisco unique blade feature and their focus on applications that can use suggests that they don’t believe – or customer feedback hasn’t led them to believe – that the other benefits of their architecture are strong enough to drive sales against HP and IBM bladeservers.

With the advent of cheap DRAM and flash, large memory servers are the next big thing. Cisco has a live one and they seem to have beat IBM and HP to market with it.

While I remain skeptical that a networking vendor can beat HP and IBM at their own game, Cisco is doing something useful by shaking up the current paradigm. Whether that redounds to Cisco’s benefit is year’s away from resolution.

Courteous comments welcome, of course.

{ 5 comments… read them below or add one }

Stuart Miniman April 2, 2009 at 2:00 pm
RC April 3, 2009 at 10:22 am

Maybe Cisco couldn’t interest anyone else in CNA over CEE?

TimC April 3, 2009 at 11:30 am

You sure the *large dimm support* isn’t just a byproduct of nehalem and the onboard memory controller?

Just because cisco announced before the chip was released doesn’t mean they’ve got mass product on anyone’s doorstep first 🙂

Robin Harris April 3, 2009 at 1:03 pm

Tim, Cisco specifically claimed it as something special they did that other blade servers won’t be able to do unless, of course, they buy MetaRam chips and integrate them on the mobo to quadruple system memory capacity (thanks, Scott).


PFC April 13, 2009 at 2:22 pm

In terms of server RAM density, when I last crunched the numbers (last fall) the increased power consumption from using more smaller DIMMs ate up all of the up-front savings within 1-3 years. So I’m not convinced cheaper chips are really a win vs. the 8GB babies. In my world, server RAM sucks more power than server CPUs, and is worth thinking carefully about.

Leave a Comment

Previous post:

Next post: