BURLINGAME, CALIF. - The devices sitting at the edges of the network are about to get a little bit smarter.
At least that's what panelists said here at the opening session of the Next Generation Network Ventures Conference.
The panelists, which included executives from Sitera C-Port, New Enterprise Associates and Intel, painted a glowing picture of the future of programmable network processors.
Programmable network processors sit between the wiring fabric in a switch and the physical layer devices. They are capable of using information from Layers 2 through 7 of the International Standards Organization's protocol stack, and process data at speeds of up to OC-48 (the 2G-bit/sec range). Networks commonly are organized according to a seven-layer architecture, each with its own protocol stack, moving from the data link layer (Layer 2) up to the application layer (Layer 7). Layer one is reserved for physical devices.
In comparison with older, hard-programmed Application Specific Integrated Circuits (ASIC), programmable network processors rely on software to define their functionality - a functionality that can be altered and changed as needed.
That is their biggest advantage, says Steve Flannery, president and CEO of Sitera, which was recently acquired by Vitesse Semiconductor.
"The value add is that network processors give you the ability to respond to new services quickly, without costly systems upgrades," Flannery says.
They also give manufacturers the ability to more rapidly implement new designs. According to Flannery, products can now be brought to market in nine to 12 months, compared to the longer 18 to 24 month time frame required for ASIC development. And since they are programmable, they will be able to stretch across a number of product lines.
These devices will reside at the edges of the network core, the metropolitan-area networks and the local loops, Flannery says.
Dr. Laurence Walker, president, CEO and founder of C-Port, which was recently acquired by Motorola, cited figures from Dataquest. Dataquest predicted that by 2003, the programmable network processor market would be worth over $800 million.
What is still in question, however is just how the processors will be programmed. Walker explained that currently an assortment of techniques are used: implementing microcode, using microcode in conjunction with 4GL languages, or programming in standard languages such as C or C++. C-Port uses the latter method.
"That makes it easy for the engineers to program," Walker says.
However much Flannery and Walker support the idea of programmable network processors, both are willing to admit that they are not for use in all situations and circumstances.
"There will always be the need for some fixed function devices, especially at the higher speeds of OC-192. It will be a couple of years before we have programmables there," Walker says.
Flannery added that by the time that happens, the core of the networks will have probably evolved into the OC-768 range.
One issue surrounding the development of programmable network processors is the question of standards. Proposals from Common Switching Interface Forum and The Common Programming Interface Forum are already being put forward as possible standards, but if and whether they will become accepted is a matter of speculation.
Mark Perry, a general partner in New Enterprise Associates, believes that customers will quickly force the creation of standards in order to ensure the levels performance and availability, but Keith Larson, director, corporate business development for Intel believes it is too early in the game to even be talking about standards. By doing so, he says it is possible to stunt the growth and development of the products.
Another issue surrounding the adoption of this technology will be the question of responsibility for the software updates.
"If you port an application, who is going to take the time to make sure that the code is hardened?" Perry asks. "Who is going to take responsibility for that code? I don't think these changes will happen overnight."
While it may not be overnight, Larson says it will happen.
"This [use of programmable network processors] will become reality, but it will take some time. I'll say two years. I think they will appear first where they are not in highly critical 24-7, five-nines systems. They will be out at the edges of the corporate networks," Larson says.
Network World, 09/15/99.
Network Processors Conference
Network World LAN Newsletter, 03/15/00.