Internet router startup Caspian Networks on Monday gave more details of the distributed platform the company says will make IP (Internet Protocol) services reliable enough for a money-back guarantee.
Going live on at least one carrier network in the second half of this year, Caspian's Apeiro switch architecture will feature software that is distributed across multiple processors and can segment IP traffic into separate "flows" that receive different treatment.
This software will run on a hardware system that can make one large router out of multiple linked platforms, allowing for scalability that keeps up with the growth of Internet traffic, said Faizel Lakhani, Caspian's vice president of network solutions, in an interview here Monday evening.
Caspian, in San Jose, California, is one of several startups that aim to solve routing-capacity shortages in the core of IP networks, a tough sell amid a glut of backbone bandwidth but critical to the future of IP service providers, according to Caspian and other vendors.
Caspian, which has stirred controversy in the past by touting forecasts of Internet traffic growth by founder Lawrence Roberts, an Internet pioneer, paints a grim picture: The capacity of traditional routers is growing roughly according to Moore's Law, or just short of it, at about 50 percent to 60 percent per year, Lakhani said. Meanwhile, Internet traffic is at least doubling every year, he said.
However, Caspian's proposition for service providers goes beyond scalability, which the company said will take its architecture to 160T bps (bits per second) in the first generation and infinitely beyond in subsequent releases. The system's capability to stay up and deliver expedited service all across the network are what could make the difference to end users.
In order to prevent failures and downtime, Caspian engineers have broken up the Apeiro's routing software into its various component tasks so they can be distributed across the processors in a platform or set of linked platforms. Each task resides on at least two processors at any given time for failover capability. Then, as the workload changes or elements fail, the system can redistribute those tasks among processors to balance the load. Software for managing the system is distributed and shifts in the same way, Lakhani said.
The reliability achieved this way helps to eliminate the need for redundant routers, cutting capital costs, he added. Caspian aims for an architecture similar in reliability to the public telephone network, he said.
"You don't see two different voice switches next to each other providing voice services, you see one," Lakhani said.
Another aspect of the Apeiro software is designed to make sure traffic that is recognized as sensitive, such as voice or video, gets the treatment it needs even after it is recognized at the edge of the network. Rather than having a set number of queues into which different types of traffic can be lined up, the Apeiro organizes different types of traffic by identifying "flows" of related packets, with no limit to the number of different flows.
Details of this feature are still vague, but Caspian says it will let carriers offer an iron-clad guarantee, which could look like a current guarantee on a Frame Relay service, stipulating a certain bandwidth available all the time. Today's IP service level agreements generally call for an average level of performance, allowing for variations that may come up at any time.
That promise has a big 'if': All the core routers in the carrier's network have to be Caspian routers to make the traffic classification work flawlessly. However, even just one Apeiro in a network can make a critical difference, Lakhani said. Currently, the cost of carrying one bit across the network probably is greater than the revenue the carrier can earn from carrying it, he said. Based on what he called conservative estimates, adding one Caspian switch to a network could tip the balance.
Networld+Interop continues through Friday.