The 'Intel Inside' of networking?
Dear Vorticians,
No, you didn't miss last week's Vortex Digest. I didn't send one. My apology.
On the Vortex Blog, I've updated a couple of stories that generated a great deal of response from you. The first was my column from January praising Google for refusing to turn over information about searches and searched sites to the federal government; the second was the piece on Comcast raising its cable prices in the face of impending competition from Verizon. You can read these updates by clicking here.
The bulk of this week's missive was actually written by a new reader of Vortex Digest, Elan Amir, who's the CEO of Bivio Networks. Elan wrote in response to my last piece on the growing influence and diversity of open source solutions. That's terrific, he says, but there's something else that could really boost the impact of open source. Vortician Amir writes:
"A colleague forwarded me your March 3 Vortex Digest and I read it with interest (and then subscribed). I agree with you that the open source model speeds innovation and technology evolution and wanted to comment on one aspect of that technology evolution that I believe may be of interest: open source, and in particular Linux, is causing a fundamental technology shift in networking and has created the inevitable combination of exciting opportunities and challenges.
"Fifteen years ago, the goal of good enterprise networking was efficient and effective connectivity. Today, we expect much more from the network. Networking has evolved from a simple switching and routing medium, requiring little processing beyond the packet header, to a policy enforcement engine providing such functions as security, application acceleration, traffic management and VoIP QoS that require deep packet inspection and real-time content processing.
"Unlike switching and routing that are easily defined in hardware, policy-based networking is inherently a software-driven function because it needs to remain adaptive to new user behavior, protocols, payload types, etc, with the ability to look anywhere in the payload of the packet to perform variable and adaptive functions. The growing range of open source solutions that you cited in your article is evidence of this evolving 'software-driven network,' which is heavily biased toward intensive packet processing and policy enforcement.
"Here's where it gets interesting. Since networking devices ultimately must operate inline to be truly effective and networks continue to get faster, in the software-driven network, an increasingly complex software-defined policy layer is now placing demands on an increasingly fast hardware connectivity layer that simply wasn't designed for it.
"This dilemma sends many a network start-up down the custom hardware and/or ASIC path to the single-function, purpose-built appliance. This incurs years of hardware and software development time, much of it redundant; millions in sunk costs before any chance of revenue; and ultimately yields a solution of limited market flexibility down the road.
"What we've been working on at Bivio for the last five years is the alternative to this model. We believe it's time to have on open, general-purpose hardware platform -- real networking hardware, not a PC -- for the software-driven network. To that end, we've combined a standard Linux-based execution environment, enhanced with a comprehensive set of common and value-added network and management features, with a unique deep-packet-handling appliance architecture that can serve as an open platform for any of the complex software applications you describe, and others that are yet to emerge.
"By providing the base networking infrastructure, such an open platform frees network application developers from costly, time-consuming ASIC development and redundant network management features, enabling them to focus on their core capabilities and values, and allowing them to enter the market faster with more innovative, scalable, and competitive solutions. It also presents a robust hardware foundation for enterprises that want to grow their own network and policy infrastructure using open-source solutions.
"As you note, 'open source platforms free developers from having to build the same fundamental product components over and over. Developers can work from a rich base of technology to add new features and functions rather than wasting time building each system from scratch." Bivio's vision is to be the next-generation network infrastructure enabling this freedom."
After thanking Vortician Amir for joining the club, I applauded his vision, but voiced some skepticism. Mainly, I've heard the story before and it hasn't made it to the big screen, it seems to me. Other companies have promised this sort of general-purpose network chip - an open development platform for all kinds of new capabilities - but it hasn't materialized. The market is still - increasingly - characterized by the inflexible, purpose-driven devices described by Vortician Amir above. So I asked him why Bivio will succeed where others haven't. His response:
"John, the main reason why existing approaches have failed in the past is that they have focused mostly on the technical aspect of the problem, rather than the overall problem, which includes not only technical but also business issues.
"On the technical side, the main issue has been that solutions have ignored the fact that the problem of high-speed deep packet inspection is multi-dimensional. You can't focus on CPU horsepower and ignore memory bandwidth. You can't focus on memory bandwidth and ignore network processing. You can't focus on network processing and ignore scalability. You can't focus on scalability and ignore flexibility. Networking is highly variant. One size never fits all, as different applications in different operating environments require varying sets of resources. As a result, it's very easy to describe and understand the idea of combining general-purpose computing and high-performance networking into one device but very hard to actually implement such a device.
"Recognizing the multi-dimensional nature of the problem, we've developed the Bivio product using a balanced architecture that attempts to eliminate any resource bottlenecks. Our product provides scalable computing, networking processing, and memory resources, while hiding all the complexity and proprietary components of the architecture under a standard Linux development and execution environment. This is in stark contrast to existing approaches that are optimized for one or two of these resources and often require proprietary development and execution environments.
"In addition to the pure hardware issues, networking requires a tremendous amount of common infrastructure. Unlike a desktop application that lives in its own world, a network device requires a significant amount of technology that sits outside of the application: management/CLI/GUI/SNMP, redundancy/HA, and remote updating and maintenance, to name a few. In addition, networking vendors spend inordinate amounts of resources on perhaps the most underestimated component of networking device development: hardware/software integration.
"Now in a world where the application is a routing stack or a forwarding algorithm that's relatively static and thin compared to the infrastructure on which it runs, you can justify developing a full product. Just as importantly, you can keep up with the required product refresh cycle. However, in a software-driven network world where the application is as complex as the infrastructure, if not more so, and where both infrastructure and application require constant advancement, development, and maintenance, the above model just breaks down from a cost, time-to-market, and ultimately competitive standpoint. In the software-driven world, network application developers are not looking for a faster chip or a bunch of technologies that replace one integration effort with another. Rather, what they are looking for is a complete solution that allows them to effectively compete at minimum cost and effort.
"If you buy into this thesis, you see that if you could provide a platform that enabled network application vendors to focus on their core competency - the application - and did not require them to compromise on any networking performance metric or feature, and in addition to that provided them with technology features and a roadmap that enables them to differentiate and innovate in ways that they couldn't do on their own, then you have something that wins where others have failed. This is what Bivio has done.
"Of course the proof is in the customer validation, and that has been going very well. We have enabled security vendors like Sourcefire, NFR Security, and Demarc to release market-leading products in a fraction of the time and cost compared to their competitors that built and maintain special-purpose devices. We're working with ISVs, large network equipment vendors, carriers, and government organizations, all of which can't believe that they can in fact get the best of both worlds - general-purpose Linux computing and custom networking hardware. As long as we continue to fulfill that vision, we feel pretty good about our chances!"
Thanks, Elan. I'll ask your fellow Vorticians what they think about the idea of a general-purpose network processor. Do you like Bivio's - or anyone's chances of becoming the Intel-inside of networking?
Bye for now.
Back to Vortex Blog
Comments
Post a comment
