To Ten Gigabit, or not to Ten Gigabit

[#1]

James Staten, principal analyst of Forrester Research, invited me to be one of the five members of a panel on data center networking economics at the Blade Systems Insight conference in April in Las Vegas.  After authoring the tagline that "Juniper will simultaneously advance the fundamentals and economics of networking, I couldn't resist.  I will attempt to bring you into the room so you too can benefit from the valuable dialog that ensued.

My fellow panelists included representatives from Cisco, Broadcom, Blade Networks and Next I/O.  James posed the question to each of us whether now is the time to adopt 10 Gigabit Ethernet (GbE). Cisco and Broadcom representatives both boldly asserted that everyone should be adopting 10 GbE network technology now either because of I/O consolidation or because of capacity requirements driven by virtualization.  That seemed rather convenient.

My guidance was to avoid the "tyranny of OR" and consider both the technical and economic needs.  Yes, 10 GbE delivers higher throughput and lower latency, but at what acquisition and operating costs?  My advice was to understand both the technical and business objectives, and to identify options addressing the technical requirements at the lowest possible TCO.  Specifically, until 10 GbE is native on server motherboards and add-on adapters are not needed, the economics of using multiple GbE ports typically more than meet I/O requirements of the data center network.  I did, however, acknowledge there are some latency-sensitive applications (e.g., trading, electronics funds transfer) that benefit from the lower latency of 10 GbE switches - both the lower clock-in time and cut-through switching techniques - that justify the higher cost.

The representatives from Blade Networks and Next I/O supported these points by noting that many of the server processors can't fully utilize 10 gigabits of bandwidth and the next generation of processors - which will have 10 GbE on the motherboard - will utilize the technology at a lower cost.

The gentleman from Cisco talked about UCS, which was expected, and James invited me to share an alternate point of view.  Based on the data I learned from the earlier case study sessions, no environment is homogenous. Juniper is a networking pure-play delivering standards-based data center solutions that allow customers to choose any server hardware, any hypervisor, any IP-based storage products.  Juniper is not in the business of creating proprietary versions of Ethernet for converged I/O or proprietary extensions for virtual switches in a hypervisor.  Juniper wants to maximize a customer's choice and flexibility so their IT strategy is just that - their's.

The Broadcom representative raised a point he observed on the high use of pass-through connectivity on blade servers.  I quickly interjected "me too, did you ask them why?".  He recited the exact lessons I learned earlier about compliance - wanting to use the data center network for consistent configuration and policy enforcement. 

A conference attendee asked "how are we seeing companies deal with the functional silos that must be broken down to deal with issues created by deploying server virtualization."  I asked the audience if anyone carries the title "data center architect."  We increasingly meet persons with this role to align the IT functions with the business requirements.  In the old days, Bob ran the network and Steve ran the servers. They may not have liked each other, but they liked the demarcation of their roles.  Now they need to set aside their opinion of each other and how they do their job, align around the higher-level business need and accept that they each need something from each other. 

James closed the panel with a question to each of us - "what is the most important value that everyone should know about 10 GbE?". 

I addressed the question with the following statements.  Use the disruption of 10 GbE as the impetus to take a hard look at data center network architectures to ensure they extract the greatest value out of game changing technologies like virtualization and evolving service oriented architectures.  Juniper's vision is to build a single-logical switch in the data center to ensure that every server has a direct flight to any other.  Today, so many switches sit between servers that data between them has to fly the equivalent of Southwest Airlines - NY to LA with five layovers. Now is the time to shift to a new architectural model that collapses layers, reduces the number of devices to lower latency and improve the performance of applications and business processes.

[#2]

Mike - excellent post. From a business perspective I agree that new technologies can be used as disruptors to take a hard look at your infrastructure. In fact, I would say that there are many factors which can be disruptors such as the case of our current economic downturn.  This caused me to take a really hard look at our cisco infrastrucure and for the first time in 10 years, explore a new vendor, and finally arrive at the decision that Juniper is the choice for us!!  Disruptors are good and innovation is even better. As IT ages - many people are getting complacent and forgetting that change is something to be embraced. 

 I also fully agree that a collapsed layered architecture is something to pursue.  I have been saying for a long time that this is necessary to reduce comlexity, cost, and allow for a more nimble environment.  Glad you are pursuing this.  All the more reason to be excited about Juniper.