Cisco Enhances Converged Networking In Switches, Servers

By pushing converged networking onto server network interface cards, Cisco is building a Unified Fabric path back to individual virtual machines.

Charles Babcock

March 31, 2011

4 Min Read
NetworkComputing logo in a gray background | NetworkComputing

Cisco Systems is making the ports on its Nexus 5500 switch capable of supporting converged networking -- either Ethernet message or Fibre Channel over Ethernet (FCoE) storage traffic.

In doing so, it's bidding to increase its claim on the design plans of those building virtualized data centers and private cloud computing operations. Plentiful server cores, servers stuffed with memory, and capacity for heavy I/O traffic are all design requirements of a virtualized server running multiple virtual machines.

The "unified" ports found on the Cisco Nexus 5598 switch means its 98 ports can support 98 virtual machine traffic channels to storage, 98 channels to the local area network, or any combination of the two. After one configuration is set up on the port, say an 8-Gigabit FCoE switch pathway to the storage area network, it can be replaced by 10-Gb Ethernet through the same port carrying communications to the LAN.

It's an extension of Cisco's entry into the blade server market, the Unified Computing System, combined with what it now calls its Unified Fabric and Unified Network Services. The ports of the 5548 or 5598 "can be either one in a dynamic way. You don't need to change cables (from SAN to LAN), the way you would in the past," said Shashi Kiran, director for data center and virtualization marketing at Cisco, in an interview before the announcement Wednesday.

Each port assumes the function assigned it by the Nexus 5548 or 5596 management console. Previously, a network administrator would have had to install two different switches and their related cabling, Kiran said.

The enhancement also reflects the distance Cisco is going to tie its line of Unified Computing System blade servers to a data center network fabric. The "unified" switching ports, as Cisco termed them, help offload traffic from a heavily virtualized server, whose multiple virtual machines are constantly generating storage and networking traffic that needs to move through the switching ports.

Since its launch 18 months ago, Cisco has tried to establish its UCS server line as possessing special capabilities for hosting virtual machines and relieving the virtualized server's I/O bottleneck. Virtualized servers are heavy consumers of memory, CPU cores, and network access. Kiran said Cisco has gained 4,000 customers for its server line through the virtual machine hosting and converged networking characteristics of UCS.

One third of the UCS servers that it sells go out the door equipped with the Cisco-recommended standard for FCoE, designated as standard Ethernet 802.1 Qbh. The standard is also supported by Broadcom, Emulex, and QLogic, but not some directly competing vendors such as Dell and HP.

In addition to giving unified ports to the 5548 and 5598, Cisco has also added the Adapter FEX and VM-FEX to its fabric extender (FEX) technologies, which expand the rack-mount server fabric's ability to cope with the traffic generated by virtual machines. Adapter FEX allows a server's 10-Gb Ethernet network interface card (NIC) to be more easily divided up among virtual machines, transparent to the applications that they are running in the virtual machines. Adapter FEX is assigned through a parent Nexus switch.

VM-FEX extends the server's switching fabric back to the individual virtual machines running beneath a physical server's hypervisor. With VM-FEX, the Unified Fabric that manages converged networking can be extended from network switches through a server's NICs to individual virtual machines. This, in turn, allows IT storage, networking, and security teams to perform their duties for each virtual machine much as they used to perform them in a physical server world. Prior to this, they had to deal with the physical server and the configuration of virtual machines before they were launched rather than with virtual machines while they were running, Kiran said.

Early virtual machine operation required each virtual machine to have its own hardware I/O devices, a NIC, and a host bus adapter to talk to communications and storage resources off the server. Both types of traffic went through the software switch in the hypervisor.

With the Unified Fabric and Unified Computing System approach, each hardware device serves multiple virtual machines, with each virtual server receiving a defined share of it. Standard 10-Gb Ethernet NICs are sufficient to serve the purpose of both host bus adapters and NICs. And the Unified Fabric reaches past the physical server's hypervisor, with its software switch, directly to the virtual machines, off-loading their I/O traffic to the network's physical devices.

HP offers its BladeSystem Matrix and related networking, including HP FlexFabric and Virtual Connect for offloading the hypervisor's virtual switch by moving converged traffic to HP networking devices. IBM is working with Juniper and other network vendors on its own approach. Dell recommends an approach using Ethernet NICs and the iSCSI storage protocol as a way to converge traffic on virtualized servers.

Cisco says it is designing surplus headroom into its approach. Its trading system switch, the Nexus 3000, not only minimizes data movement latencies but also includes that ability to handle 40-Gb-per-second traffic and will one day handle 100-Gbps traffic.

About the Author

SUBSCRIBE TO OUR NEWSLETTER
Stay informed! Sign up to get expert advice and insight delivered direct to your inbox

You May Also Like


More Insights