Today Cisco announced the NEXUS switch family, a remarkable redefinition of data center switching. Over the last five years the data center has become more and more of an IT focus driven by these forces:
1. The increased deployment and use of shared, server-based applications and services,
2. The desire to improve IT cost efficiency (through consolidation),
3. The need to more carefully govern and control IT activities (regulatory compliance) and
4. The need to protect sensitive and private data.
The increased focus on the data center has in turn made it a fertile ground for innovation and yielded technology-enabled breakthroughs such as data center virtualization. As data center technologies and architectures evolved the network stood out as much as a limitation as an enabler in ways that seemed difficult to fix with just improved features.
From this perspective we can see that what Cisco has done with Nexus is an impressive systematic improvement to data center switching. Some of the elements we have seen before (what Cisco calls switch “virtualization” for example). We knew that Cisco was working with Nuova to improve Ethernet so that it could serve as a transport for fiber channel SAN protocols. We had already seen in the Catalyst offering how Cisco has evolved the product line for customer investment protection assuring that essentially all the components in the switch can be upgraded independently -- new control processors, new data backplane, and of course new line cards -- and that the old line cards had firmware flexibility to adapt to evolution in other parts of the switch. We knew that Cisco could and liked to do high-speed custom silicon (a great barrier to entry). NEXUS has all of that built into what is really a completely new network system for data center applications.
NEXUS represents a large and expensive product development effort for Cisco which we won’t try to describe in any real detail, but here are the key aspects:
1. Evolve switching so that Fiber Channel SAN traffic can use the same network fabric. NEXUS not only carries LAN and SAN traffic over the same switch but can use a single port for both. Converging these networks simplifies the data center infrastructure, reduces CapEX and OpEx and according to Cisco’s numbers saves an amazing amount of power when you consider the reduced number of server adaptors required.
2. Create a data center network architecture where all ports are equal and minimize the need for any physical network moves and changes or re-cabling while supporting the IT agility inherent in a virtualized data center (the ability to move applications between servers at will).
3. Reduce CapEx while providing very high availability. This is where the switch virtualization yields high returns by eliminating the need for “standby” devices and using them in the network fulltime.
4. Improve network availability by greatly improving the network re-convergence time after failure. Cisco makes heavy use of Level 2 coordination of the devices thereby avoiding the recovery delays that come with traditional spanning tree algorithms and also thereby flattening the topology toward the goals of having all ports equal for all tasks.
The net result over time is a data center network that is faster, more reliable and cheaper, hence what we think of a real step forward. Along the way the feeds and speeds improve a lot. Having enough backplane capacity to assure that the backplane is never the limitation was key to the integration of FC SAN traffic. The switch uses a new operating system derived from the SAN-OS used in the MDS 9000 SAN switching products. FC requires assured delivery which is quite different from traditional packet best effort delivery. Since a lot of the features required in WAN routing don’t need to be supported in the data center it made sense to move to an O/S platform with these key SAN primitives.
Cisco says that they set about rethinking data center switching when they began this effort quite a few years ago. Typically “rethinking” is added by the marketing people when a project is done but in this case we think it’s probably true. The kind of wholesale systematic redesign is something only Cisco can do in enterprise networking (remember they own 70%+ market share to their nearest competitors 7%). Most dominating companies get lethargic and spend more time resting on their laurels and improving the bottom line a little by constraining R&D spending. We’re delighted that Cisco is still driving progress so aggressively.
Comments