Cisco is working on disaggregating its Unified Computing System (UCS), dividing the design into modular pieces that can be upgraded separately.

The company started down this path in 2014, with the UCS-M family. Today, it's announcing a further step with the UCS-S, targeting unified storage and computing.

"This is a view into where we're heading with UCS at large: increased disaggregation of subsystems, using our System Link technology to allow for much more composable infrastructure, more agile infrastructure," says Todd Brannon, managing director for UCS.

That kind of modularity will become important "especially as we see things like containers come down the road," he says.

In disaggregating UCS, Cisco is taking the same approach that Intel has developed with its Rack Scale Design. (That architecture is used in the HDS 8000 developed by Ericsson, which has a broad partnership with Cisco.)

The idea is that different elements, such as flash drives for storage, or CPUs for computing, can be upgraded independently of one another. This means storage or CPUs can be upgraded independently to keep pace with the latest versions of either technology. It also means the ratio of storage to CPUs can vary depending on the specific application being served.

The first S-series member, the UCS S3260, targets on-premises data storage for operations that are heavily using the cloud.

Those terms clash a little bit, but the scenario Cisco has in mind is an operation that's collecting massive amounts of data. If all that data gets stored in the cloud, there's a higher likelihood that it's just going to sit there, unused. And while it's easy to think that storage is cheap, it adds up if you're talking about hundreds of terabytes sitting on Amazon Web Services (AWS), Brannon says.

The S3260 and its successors would create the option of not only keeping some of the data on-premises, but also keeping it right next to some compute power for performing analytics. The slogan is that Cisco is helping customers "unstore and activate" data, Brannon says.

The S3260 is 4 rack units tall and can hold of 600 TBytes of storage. It's due to be available starting Nov. 7.

The disaggregated UCS isn't the same as hyperconverged infrastructure, in which pooled CPUs and storage are managed by a hypervisor. (This is what companies such as Nutanix, Pivot3, and SimpliVity do.)

It's also not quite the same thing as composable infrastructure — the step beyond HCI that Hewlett Packard Enterprise (HPE) has been talking about. That scenario creates pools of compute and storage that can be aggregated regardless of location, so that applications can spontaneously call up the resources they need.

Nice idea, but that's not what UCS is going for, Brannan says.

"A lot of the conversation that's coming around composability — we think a lot of customers are going to see it as added complexity," he says. UCS-S, by contrast, is meant to be more like "automatic transmission on a car: They don't know how it works, and they don't want to."