Cisco preps Nexus switches for third-party blades
You can beat them, and you can join them
Networking giant and server wannabe Cisco Systems says it will be working with third-party blade server makers to create a version of its Nexus family of switches that tuck inside non-Cisco blades.
The details are a bit sketchy (intentionally so on the part of Cisco). But today, during a Webcast discussing its unified server-storage fabric efforts, Ed Chapman, vice president of product management at Cisco's Server Access and Virtualization Group, said that the company would be soon rolling out a variant of the Nexus 5000 top-of-rack converged switch that would fit inside of a third-party blade chassis. Chapman called this the Nexus 4000, and he didn't say much else about the device. So forget about getting feeds and speeds.
The basic idea must be to shrink the Nexus 5000 so it can handle the 10, 14, or 16 blades in a chassis offered by the likes of Hewlett-Packard, IBM, Dell, and others. The Nexus 5000 switch, of course, is the switch that handles server and storage traffic (using the Fibre Channel over Ethernet, or FCoE, protocol) all on the same device. A variant of this Nexus 5000, called the Nexus UCS 6100 Series Fabric Interconnect, adds the Cisco UCS Manager for the "California" Unified Computing System, which manages blade servers and switching between servers and the outside world and their access to storage.
By shrinking the Nexus 5000 10 Gigabit Ethernet switches, Cisco is able to stay in the game with the tier one server makers it has alienated a bit with the California boxes, and it will presumably allow the blade server makers to interface with the Cisco switches but maintain control of the management of the blades and switches inside the chassis - which is what the server makers really want.
What Cisco and its partners will be touting the ability of the Nexus 4000 to do plain-old high bandwidth, low-latency computing. Chapman said that the unified switch ASIC used in the Nexus 4000 would be able to deliver around a latency of around 1.6 microseconds, which is fast enough for high performance computing and other latency sensitive jobs. (But by no means the lowest latency switches available today running at 10 GE speeds, mind you.)
The Nexus 4000 have 10 GE ports for both uplinks and downlinks and will use the same Nexus OS operating system used in the MDS 9000 Fibre Channel switches and the Nexus 7000 end of row, Nexus 5000 top of rack, and Nexus 6100 California switches. It is also expected to have fault tolerant features and work seamless with the Nexus 1000V virtual switch that Cisco has created in conjunction with VMware to run inside of an ESX Server 4.0 virtual machine.
According to a Cisco spokesperson, Cisco will be working with the blade server makers to cook up Nexus 4000 switches that are paired with their specific blade servers and chassis, and server makers are expected to make their own announcements about their plans in the coming weeks.
On the Webcast, which was mostly a cheerleading session for unified fabrics and unified computing, Rajiv Ramaswami, vice president and general manager of Cisco's Data Center Switching Technology Group, reminded everyone that Cisco did not think for a second that Fibre Channel connectivity itself was dead. Ramaswami pointed out that IBM introduced ESCON channels for mainframes back in 1988, and many mainframes still use it. Cisco expects some customers to need (or at least want) unconverged Fibre Channel switches for years to come.
To that end, Ramaswami said that Cisco is working on 16 Gb/sec Fibre Channel links for its MDS 9000 family of switches. The Nexus 5000 will soon be upgraded from 4 Gb/sec Fibre Channel links out to storage arrays to 8 Gb/sec modules. The exact timing of the faster ports was not revealed. Ramaswami added that Cisco was cooking up Fibre Channel over Ethernet modules for the MDS 9000 and Nexus 7000 switches, allowing for end-to-end use of the FCoE protocol for companies that want to do that, and would be cooking up legacy support for 100 Mbit and 1 Mbit links in the fabric extenders for the Nexus 5000 switches so they could link to older server kit that does not have Gigabit Ethernet ports. The Nexus 5000 switch will also get a 10 GE FCoE fabric extender, and the Nexus 7000 is slated for some kind of fabric extender support as well (but the speeds were not revealed).
Finally, Ramaswami said that VMware and Cisco were working on the development and testing of long-distance VMotion for the ESX Server 4.0 hypervisor and its related vSphere 4.0 tools. ®
Sponsored: Benefits from the lessons learned in HPC