NGN Ventures: Rethinking the Internet Traffic Model - Beyond TCP/IP

The bubble era's brute force solutions of bandwidth and people does not scale and cannot work any longer, said Doug Brent, CEO of Packet Design, at the NGN Ventures conference. A smarter system is needed, and Packet Design believes that route analytics based on control plane analysis can provide it. Network layer analysis today is highly limited. Route analytics provides up to date information accuracy, root cause analysis, and scalability to support large networks. It offers a forensically accurate view of a network, enabling the diagnosis and solution of specific problems. While IP will continue to be the unifying force in networks, the industry needs to make IP networks better, not just faster.



We have to deal with IP as it is. It would be nice to improve it, but can't at this point, commented Larry Roberts, CTO of Caspian Networks. Scalability, however, can be addressed by changing routing to a system that is not fixed. Roberts believes there are several misconceptions about state-based IP routing. To those who say it cannot scale, he responds that state-based routing permits a more efficient switching fabric, the routing of individual flows and fast recovery. To those who argue that mainting state is expensive due to the system memory needed, Roberts says that it can be solved with inexpensive DDRAM, which works fine in a state-based system. Caspian's platform routes flows rather than packets over the network core, supports all aspects of IP, MPLS, ATM, TDM and FR for any flow, and can be used to provide QoS and bandwidth guarantees.


Digital Fountain is rethinking the method of transferring data over the WAN, said President and CEO Cliff Meltzer. The core problem is that speed in the WAN and core is of limited use if there is no predictability. TCP throughput falls as round trip time increases. Since TCP requires the ordering of packets, its performance dramatically degrades as packet loss increases. Digital Fountain has developed a distance insensitive solution to data transfer. Appliances at the origination and termination points of the network create and decode meta content packets that are a precise mathematical "recipe" for the original data, which can be of any type or format. Packets do not have to be delivered in any specific order. It is only the number of packets delivered that is tracked. As a result, throughput declines only by the exact rate of packet loss in the transfer of Digital Fountain data. Digital Fountain is targeting three enterprise markets: global organizations that need to share data across geographically diverse locations; the media and entertainment industry, which is undergoing a radical transition from analog tape to digital storage; and collaborative engineering groups. Meltzer said that the military has also been interested in its technology.


Up to 90% of network traffic is repetitive, according to Amit Singh, CTO of Peribit Networks. To optimize enterprise WAN application performance, Peribit decided not to address how data is delivered, but to modify the data itself. Their network optimization appliance encodes entire data streams, leading to fewer bytes and fewer packets, resulting in more effective use of capacity. Singh said that existing WAN compression schemes are based on LZ compression, which does not offer offers advantages in high speed networks. Peribit uses Molecular Sequence Reduction, based on concepts identified in DNA pattern identification. It offers high speed, scalable, real time repetition identification anywhere in the data stream. Peribit's sequence reduction engine typically cuts stream sizes by 75%. The company's technology requires no routing changes, works with private IP and VPN networks, and offers QoS and bandwidth management.


New IP models offer some of the most interesting investment prospects, according to Jim Goetz, General Partner at Accel. As an investor, he finds application specific opportunities more attractive today than efforts to solve everything that is wrong with the network. He also feels that good opportunities exist in specific areas that Cisco does not control, and where there are broadly deployed technologies with common problems. For example, Microsoft Exchange is ubiquitous, but it causes many headaches. If a solution targets this problem, it could be a great opportunity. Goetz believes that there are several challenges for appliance vendors. To survive and make itself truly valuable to its customers, a company must offer a greater value proposition than one solution delivered over a single appliance. Given the difficulty and expense of building a distribution network, he also believes that a company needs to sell at least $1 million of products and services per customer per year for long term growth. Otherwise, small sales per customer will be eaten up in sales and marketing expenses. Goetz also said that start-up frugality is back in vogue, and the new investment hurdle may be $25 million for a startup to reach breakeven.


A new protocol will not be successful, said Shirish Sathaye, General Partner at Matrix Partners, but a point to point targeted solution can be. Speeding data transfer is of true value to many businesses, and solutions that address this issue will be bought. He feels that an appliance on each side of the public network is a good way for an enterprise to achieve this. It allows them to improve Internet performance in a way they can afford and control. He does not expect many successes from software-only companies in data transport. Even if a company's skill is software, customers will want to buy the technology in its own box.


Co-chairman John McQuillan observed that companies which take a "rifle shot" approach to specific needs may have a better chance to succeed in this environment than those which confront huge the holistic solutions, but the ultimate growth and size potential for these companies is not very large and they cannot take in as much investment.