News Stay informed about the latest enterprise technology news and product updates.

Cisco FabricPath turns to 10 GbE networks for virtual servers

Technology promises virtual server migration across domains, but the devil's in the details -- and the price.

Cisco Systems claims the new FabricPath protocol to be added to its Nexus operating system (NX-OS) in the third quarter will facilitate virtual server migration across domains, boost scalability and lower latency in data centers.

For more on Cisco's Unified Computing System
New Cisco Unified Computing System faces tough sell

IT pros ponder Cisco's Unified Computing System

Cisco Unified Computing System vs. VMware vSphere 4

Defining 'unified computing systems'

Virtualization experts call the plan, announced at Cisco Live in Orlando, Fla., ambitious -- and perhaps too pricey for most shops. The protocol is also designed to integrate with Cisco Systems Inc.'s Unified Computing System (UCS), which data center managers have eyed warily given its price and potential for vendor lock-in.

The FabricPath basics
The proposed FabricPath protocol is based on the Transparent Interconnection of Lots of Links (TRILL) standard currently from the Internet Engineering Task Force (IETF) The protocol is a successor to the spanning tree protocol, the traditional method of routing traffic between separate Layer 2 domains.

Some people want to wait until UCS has a few years under its belt.
Chris House,
senior network analystThe MetroHealthSystem

This method relies on "blocking" links to isolate paths and avoid creating network loops as the network tries to resolve a destination for data across domains, slowing the network and making it less scalable, according to Shashi Kiran, Cisco's director of corporate marketing for data center networking.

A "flattened" layer 2 domain is a key ingredient in some of the more advanced architectures proposed for virtual server deployment, including the distance VMotion concept demonstrated by VMware and Cisco at VMworld 2009. Distance VMotion would allow workloads to be moved between geographically separated data centers in the same way they are moved between physical servers within a single data center today.

FabricPath is also a successor to Cisco's vPC and FEXlink protocols, which allow for dual-active paths. FabricPath allows up to 16 active paths simultaneously, and Cisco claims it can now scale up to 2,048 servers connected by 10 Gigabit Ethernet (GbE) at full line rate without blocking. Another component of the architecture that was introduced in February, Overlay Transport Virtualization (OTV), provides a tunneling mechanism to make it appear to applications that networks at separate data centers are part of one large domain. FabricPath will be available as a module for NX-OS licensed at $25,000 per switch chassis.

Users eye Cisco's vision, weigh practicality
Chris House, a senior network analyst at The MetroHealth System, a hospital network based in Wyoming, Mich., said the company's networking team is gathering information on Cisco's Unified Computing System but has already found cost a gating factor. "It has to cost the same as a competitive HP solution," said House. Currently, MetroHealth uses HP equipment extensively, from servers to storage. "Some people want to wait until UCS has a few years under its belt."

In the meantime, House said, there isn't necessarily a need to wait for Cisco to do cross-data center routing of network traffic. "We currently have the same network in both data centers: flat layer 2," he said. "Our network team wants to move up to layer 3 and do routing everywhere, but still allow us to use the same IP addressing scheme … in each site, so something like what Cisco has announced would be very beneficial. Though I believe our network team has come up with other ways to do this using our existing network equipment [and] various tunneling protocols."

The price and scale of FabricPath -- Cisco's reference customers are HPC shops and T-Systems, a large service provider subsidiary of Deutsche Telekom -- may also put it beyond the reach of most enterprises, said Rick Vanover, an IT infrastructure manager at a large Midwestern financial services firm.

"It will surely be a boon to very large enterprises with a substantial infrastructure that includes virtualization. The issue I see is that many large organizations are in between product lifecycles and can't quite rip everything out to take advantage of all of this new stuff. Also, the medium and smaller environments are totally left in the dust, it seems," he said.

The CTO for a large solution integrator on the East Coast said practicality is the rub when it comes to TRILL standards, since few data centers have a completely homogenous infrastructure. For example, "at the end of the day, Brocade rules 80 percent of the storage switch market. There needs to be interoperability, and it needs to be documented. When you go into a data center with a mixed environment, what do you do?" he said.

In their rush to capture territory in the converged data center, vendors like Cisco and HP "have been giving us 'you're either with us or against us'," he continued. "We can understand both architectures, but the value we bring as partners is a day-to-day understanding of how to manage the junctions [between technologies] and bring the pieces together in a practical way. Clarity on where those junctions are from an architecture rather than a marketechture standpoint is what we need from Cisco."

Instead, though, Zeus Kerravala, a distinguished research fellow at the Yankee Group, said the approach to standards with FabricPath is "a very typical go-to-market for Cisco. You can look across market transitions in other areas, and their typical proprietary way is to do things better than the standard before the standard is ratified, gain an early-mover advantage, and then support the standard when it comes out. By then, customers that have adopted the pre-standard product feel going to the standard is a downgrade."

That said, the TRILL standard, and the things it enables, like FabricPath, are needed, Kerravala said, and Cisco already owns a large chunk of the Ethernet switching space already. "They have enough share and the customer base to be able to do this."

Broader WAN optimization and new services
At Cisco Live, the company also added new switching hardware to its portfolio that it says will support FabricPath in the form of a new line card for the Nexus 7000 series switch and Catalyst 4948-E. The Nexus 7000 F-Series module will scale up to 512 10 GbE ports per system and can support both Gigabit Ethernet (GbE) and 10 GbE with auto-sensing features. The 4948-E supports up to four 10GbE uplinks or 48 GbE ports. The 4948 without the "E" supported up to two 10 GbE uplinks. The 4948 also supports IPv6 in hardware, rather than software, offers a larger buffer and call-home features. The switches are available now at $35,000 list price for the F-Series module and $10,495 for the new Catalyst.

Cisco also updated its Wide Area Application Services (WAAS) WAN optimization product. Version 4.2 is now supported on Cisco's branch-office router, which allows services to be deployed remotely onto 'blank' hardware modules and managed centrally. WAAS 4.2 can also optimize http and https protocols for Software as a Service services and connections to the public cloud. It can also now be deployed on Windows 2008 R2 servers. WAAS 4.2 is available now with starting prices for various product bundles ranging from $2,500 to $8,100.

Finally, Cisco unveiled new professional services offerings designed to push IT into the cloud. New Cloud Enablement Services will include a Cloud Strategy Service, Cloud Planning and Design Service, and a Cloud Implementation Service. "Cloud Enablement Services sounds interesting," said MetroHealth's House. "I have maintained that enterprises would benefit from cloud guidance from trusted VARs or other folks such as Cisco."

Beth Pariseau is a senior news writer for Write to her at

Dig Deeper on Server hardware and virtualization

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.