Older: 
Cisco VSS….too little, too late?
As I discussed in a previous set of blog postings, delivering redundancy is a critical path to assuring overall system availability. For a number of years, Nortel has deployed a patented implementation called Split Multi Link Trunking (SMLT) that uses the IEEE 802 standard in a configuration that is active-active with incredibly low switchover time. When compared with traditional techniques, such as spanning tree or routing, this technique has been judged vastly superior by both our customers and by independent testing organizations such as the Tolly Group. Nortel has deployed this technology for over six years to deliver effective Switch Clustering in a number of Ethernet Switching platforms. This is important, as the Nortel Switch Clustering means small and medium enterprise customers don't have to compromise on resiliency as Nortel offers a range of products that meet their architecture needs.
The recent Cisco announcement of their long awaited Virtual Switching System for the Catalyst 6500 LAN Switch is a dramatic endorsement that delivering redundancy is a very real networking challenge. Without a form of technology guaranteeing active-active redundancy, networks are mired in the 90s. Cisco’s VSS announcement clearly confirms there are significant deficiencies in their current offers based on various combinations of spanning tree and routing. However, the announcement also seems to expose serious flaws in Cisco’s latest attempt to tackle active-active redundancy. I thought it would be interesting to contrast the new Cisco VSS with the proven Nortel SMLT solution and see what the critical differences are. It is important to note that all of my comments are based on publicly available data, and any erroneous interpretations are not intentional, but may be a result of the limited information Cisco has chosen to release about it's VSS system.
From the announcements, it would appear that Virtual Switching System (VSS) is only available on the top-of-the-range, fully-loaded Catalyst 6500 Switch. Fitted with the Supervisor 720-10GE module; the Virtual Switch Link (VSL) must be made with 10G Ethernet connections, using either the 10G ports on the Sup720 or those on the X6708-10GE with DFC3/CXL Interface module. Further, in order for the system to work, it appears all interfaces in the chassis must have the latest DFC functionality installed, potentially causing an expensive upgrade for any older modules. SMLT, on the other hand, works across all the modules in a Nortel 8600 and does not require card level upgrades or specialized interfaces.
The Cisco VSS seems to essentially turn two switches into a single chassis by connecting the control processors together across the VSL. While in some cases this may have limited advantages in configuration and management, it would appear to have serious implementation drawbacks. Between two switches operating as a VSS pair, there can only be 2 Sup720 modules. This means that when there is a failure or other reason to take one of the chassis down (for re-configuration for example), the remaining chassis will only have a single processor, exposing it to failure issues. In the Nortel implementation, each chassis can have both a primary and secondary processor, essentially doubling the overall redundancy at minimal cost. This leads to two other questions that can only be answered with testing; does having a single Sup720 supporting 2 chassis over a VSL result in reduced operating capacity or other performance issues and does the encapsulation techniques used on the VSL result in performance issues in latency or scale during large scale failures? The Nortel Switch Clustering using SMLT has proven that it excels in both scaling and transitions as it enables dual active processors in each chassis and does not use proprietary encapsulation on the inter-switch links.
Finally, it is not clear if or how the VSS can interoperate with open MLT access devices such as wiring closet switches and server NICs. A critical strength of the SMLT solution is seamless integration with standards based MLT devices. The SMLT solutions do not require special software and have proven interoperability with a number of other vendors wiring closet switches, including Cisco as well as a wide variety of NICs in servers. In fact, Nortel SMLT has been identified by VMWare as the optimal solution for multi-connections in a VMWare virtualization environment. While it appears that the VSS uses software in the access devices for some levels of functionality, it is not clear whether or how it performs in an open multi-vendor environment and how this limits functionality when the devices are not Cisco..
All of this means that network architects now have to decide which Cisco design methodology to use – VSS, L3 or traditional Spanning Tree– and very serious budget considerations must come into play. The promotion of yet another non Standard-based approach means further concern about customer choice.
I think it is time for a reality-check. Do we really need more Cisco proprietary technology to make our networks and businesses more complex? Clearly the answer is ‘no’, and Cisco’s move is in direct conflict with recent discussions within the Analyst, Media, and User communities: Cisco’s solutions are increasingly recognised as being expensive and not worth the price delta. I believe users should actively avoid the use of proprietary solutions that lead to single vendor lock-in.
Nortel has established a proven track record with our SMLT and Switch Clustering architecture for more than six years now, delivering deterministic application reliability and availability, all with a fully open and tested interoperable model. The real beauty of Nortel’s solution is that the principle business benefits are consistently delivered across the portfolio (both in multiple product and platforms as well as a variety of interface speeds), and appropriately scaled and priced for deployment from corporate headquarters to regional offices. This enables businesses to enjoy consistent service delivery and applications reliability, further enhancing application usage, performance, and value.
However, I would like to thank Cisco for endorsing the need for the technology leadership that Nortel has demonstrated and for recognizing some of the issues with their previously proposed implementations.
[…] as well as a penchant for trying to present attempts at catch-up as innovation - see my post on VSS. In the VSS case, the Cisco technology is proprietary and appears to be a pathetic effort to […]
February 21st, 2008 at 11:59 am from Enterprise Technology » Blog Archive » Merchant Silicon - Benefit or Bane?????