Cloud Networking Reflections

Jayshree Ullal

Every year I like to reflect on my previous years predictions and look at how well they mapped to what actually happened. Once again, that time has come – so let’s take a walk down the 2011 memory lane to judge how I fared!

Prediction #1: The rise in dense virtualization is pushing the scale of cloud networking.

Evaluation #1: True.

VMWorld 2011 experienced a transformation from prior years. This year really saw the maturity of virtualization from being a server-only consideration to one that is now network-wide. Until now, administrators mostly dealt with just the server side, but with new emerging specifications, for example, VXLAN which was co-authored by Arista and VMware (Arista has also co-authored a similar specification with Microsoft for NV-GRE), one can transcend network boundaries at both L2 and L3 and avoid building separate virtual and physical switch islands. Arista’s award-winning VM Tracer is the industry’s finest network virtualization software that works across virtual, physical and cloud networks, interoperating seamlessly with VMware’s vSphere and vCloud Director.

Prediction #2: “Fabric” has become the marketing buzzword in switching architectures from vendors trying to distinguish themselves.

Evaluation #2: Half-True.

The industry is still confused by these vendor-hyped Fabric marketectures. They must be clarified better in 2012. An internal fabric is always required for all switches and defines maximum capacity but vendors are now pushing their own fabric “extensions” outside the box.

QFabric, Fabric Path, Fabric Extenders and Fabric One from leading companies all have one common flaw. They introduce proprietary mechanisms between their switches making multivendor interoperability and debugging difficult. All claim to improve scale and offer a single point of management. Yet many analysts, customers and I wonder why we need these proprietary lock-ins? Are these really inventions needed for scale or management?

The beauty of Arista’s approach is we can scale and manage order of magnitude (10-20 times better) with standards. I fail to understand the need for vendor-specific proprietary tags for active multipathing when standards-based MLAG at Layer 2 or ECMP at Layer 3 (and future TRILL) resolves the challenges of scale in cloud networks. Arista’s CloudVision also demonstrates that a single point of management does not require proprietary inventions and can be achieved by utilizing standards to support thousands of switches.

Prediction #3: OpenFlow and Software Defined Networking (SDN) will change the way networking is deployed.

Evaluation #3: Half true.

The hype on OpenFlow is presently at a feverish pitch, but use cases are still emerging. At Arista we have taken a pragmatic approach to SDCN (Software Defined Cloud Networking) to ensure that our modern and extensible OS, EOS, is controller friendly and can work with multiple APIs, be they Netconf, XMPP, CLIs, or future OpenStack and OpenFlow interfaces. We believe that end users will not replace today’s successful network, but may augment the network control plane with appropriate additional functionality to improve performance and scalability. Arista EOS is ideally designed for both L2/L3/L4 and SDCN models.

Prediction #4: Commercially available silicon offers significant power, performance and scale benefits over traditional ASIC designs.

Evaluation #4: Very true.

The industry is consequently witnessing an increased adoption of commercially available merchant silicon in data center switches.

Arista has been a consistent and vocal fan of commercially available silicon. We find most in-house ASICs are two or three generations behind their modern commercially available silicon counterparts. Many silicon choices from suppliers such as Broadcom, Intel (through its acquisition of Fulcrum Microsystems), and Marvell are available today. In most cases, chip validation is more advanced and the geometry of these chips is smaller in footprint with lower power and time-to-market advantages. Perhaps the greatest validation of this premise was Cisco’s own adoption of merchant silicon from Broadcom into a specific switch product in 2011. Time will tell if this is a one-off act or the de facto realization of a commercial reality.

Prediction #5: FCoE adoption has been slow and not lived up to its marketing hype.

Evaluation #5: True.

FCoE (Fiber Channel over Ethernet) is still a niche technology in terms of market acceptance, with numbers estimated at $100-200M in switch sales. Meanwhile the $3B Fiber Channel SAN market has been upgrading from 8 to 16 to 32G in high-end enterprises, while 10GbE, already a multibillion dollar market, is becoming the storage technology of choice for NAS, iSCSI and direct-attached storage such as that used in Hadoop clusters. A key criterion for using 10GbE in storage applications is having switches with adequate packet buffering that can cope with speed mismatches to avoid packet loss and balance performance, which is a hallmark of the Arista 7000 family of switches. Furthermore, by working closely with our storage partners we make it possible for any storage protocol to work optimally over 10GbE – not just FCoE.

The Bottom Line

In summary, I think it’s safe to say that I didn’t score perfectly in my five predictions. I would self-evaluate it at 4/5. The migration trend to 10GbE in key verticals is well underway. At Arista Networks, we are proud to be a market pioneer and leader with 1000+ customers deployed in 1000 days. I want to take this opportunity to sincerely thank all my readers, and customers for supporting Arista’s innovations. We wish you all Happy Holidays and a Happy New Year!

As always, I welcome your comments at