One of the highlights of the Gestalt IT tour was a half day on Cisco’s UCS and associated products. But this was the real deal: an experienced and technical Cisco presenter going deep for a crowd of skeptical IT pros. Digging into the details would reveal the essential value proposition of UCS.

I’ll summarize what I heard first. This isn’t a transcription and there is interpretation and interpolation. If there is something wrong in the descriptions please comment so I can update. There are links to Cisco sources for more detail.

The UCS case
UCS is focused on reducing operating expenses not capital expenses as OPEX is the rising cost. It is proprietary software and support services that breed cost.

Customers are caught on a treadmill, where vendors:

  • Simplify by adding layers and offering services to enable the simplification
  • Which results in a complex management stack and high costs
  • Enterprises can’t easily scale because of legacy systems, which drives application costs higher
  • Management complexity is driving vendor revenues

Sounds something like a StorageMojo critique. Bravo!

The UCS solution is to move to a private cloud powered by VCE – VMware, Cisco and EMC – and other vendors as appropriate. An application-centric unified fabric that ties network and compute resource together under centralized control.

The building blocks of the UCS are:

  • UCS manager – the device manager
  • UCS fabric interconnect – 20 & 40 port FCoE switches
  • UCS fabric manager is the management tool for storage networking across all Cisco SAN and unified fabrics
  • Fabric extenders connect the UCS blade chassis to the switch and simplify cabling, management and diagnostics

All these components are designed to work with both physical and virtual resources.

UCS components
The basic Cisco components of the UCS are:

  • UCS manager: Cisco UCS Manager implements policy-based management of the server and network resources. Network, storage, and server administrators all create service profiles, allowing the manager to configure the servers, adapters, and fabric extenders and appropriate isolation, quality of service (QoS), and uplink connectivity. It also provides APIs for integration with existing data center systems management tools. An XML interface allows the system to be monitored or configured by upper-level systems management tools.
  • UCS fabric interconnect: Networking and management for attached blades and chassis with 10 GigE and FCoE. All attached blades are part of a single management domain. Deployed in redundant pairs, the 20-port and the 40-port offer centralized management with Cisco UCS Manager software and virtual machine optimized services with the support for VN-Link.
  • Cisco Fabric Manager: manages storage networking across all Cisco SAN and unified fabrics with control of FC and FCoE. Offers unified discovery of all Cisco Data Center 3.0 devices aa well as task automation and reporting. Enables IT to optimize for the quality-of-service (QoS) levels, performance monitoring, federated reporting, troubleshooting tools, discovery and configuration automation.
  • Fabric extenders: connect the fabric to the blade server enclosure, with 10 Gigabit Ethernet connections and simplifying diagnostics, cabling, and management. The fabric extender is similar to a distributed line card and also manages the chassis environment (the power supply, fans and blades) so separate chassis management modules are not required. Each UCS chassis can support up to two fabric extenders for redundancy.

Here is a simplified figure of the components, courtesy of Cisco.

Cisco value-add
The presentation noted several unique Cisco features:

  • Memory extension. Cisco blade servers support up to 48 DIMMs due to a custom mux/demux chip they developed. Enables a 96 GB server using low-cost 2 GB DIMMs.
  • Hypervisor bypass. Bypass the softswitch to go direct to the NIC using Single Root I/O Virtualization – which is part of the PCI spec, not a Cisco exclusive.
  • Exceptional automation. Blades are highly stateful – MAC addresses, BIOS settings, vLANs – are put in an XML file so you configure the blade once and apply forever with compatible blades. Settings can be applied to either a blade or a particular slot. Removes much sysadmin drudgery.

VMware/Cisco/EMC: VCE
That was just the Cisco-owned UCS story. But VMware and storage are needed to create a virtualized infrastructure. Enter, in this instantiation, EMC and the Vblock. Cisco is also working with NetApp and probably others,

A Vblock is an engineered, tested, supported and validated package of components from the 3 vendors. You buy the package – which has some configuration flexibility – and you get a single support group, not finger pointing between 3 companies.

This is supposed to make for rapid implementation of new infrastructure along with the management advantages of UCS. Acadia, the new services company the 3 have put together, provides Build, Operate & Transfer (BOT – a new TLA?) services.

Sounds good. I can’t recall a similar level of advertised integration among 3 major vendors before.

The StorageMojo take
I’m just as mystified as ever about what they’re thinking.

The looming question is: do enough customers want to buy “unified” systems? This is, after all, only a distributed mainframe – and the mainframe spending percentage has been shrinking for decades.

Consolidation is inevitable: IT has standardized on a few platforms, and can standardize on a few suppliers. But do those suppliers need to be vertically integrated?

Only when it makes sense. Oracle/Sun has the better argument: when you know exactly what you want from your database, we’ll sell you an integrated appliance that will do exactly that. And it’s fine if you roll your own.

But those are industry-wide issues. There are UCS/VCE specific issue as well:

  • Cost. All the integration work among 3 different companies costs money. They aren’t replacing existing costs – they are adding costs. Without, in theory, charging more.
  • Lock-in. UCS/Vblock is, effectively, a mainframe with a network backplane.
  • Barriers to entry. Are there any? Cisco flagged hypervisor bypass and large memory support as unique value-add – and neither seems any more than a medium-term advantage.
  • BOT? Build, Operate, Transfer. In theory Vblocks are easier and faster to install and manage. But customers are asking that Acadia BOT their new Vblocks. The customer benefit over current integrator practice? Lower BOT costs? Or?
  • Price. The 3 most expensive IT vendors banding together?
  • Longevity. Industry “partnerships” don’t have a good record of long-term success. Each of these companies has its own competitive stresses and financial imperatives, and while the stars may be aligned today, where will they be in 3 years? Unless Cisco is piloting an eventual takeover.

The enterprise IT industry is consolidating. HP, the world’s largest computer company, appears strong but is vulnerable – or at least John Chambers, Cisco’s CEO, thinks so.

Cisco, dominating network switches, needs new worlds to conquer. Large switches have been specialized blade servers – CPU and I/O – for decades, so why not take the next step?

But Cisco is responding not to customer demand, but to Google and Amazon. Their vast commodity infrastructures, linked by – horrors! – cheap unmanaged switches is Cisco’s nightmare. If CFOs understood that much of IT could be migrated to that model over the next decade, Cisco’s margins and influence would be devastated.

Creative destruction, indeed!

Courteous comments welcome, of course.