High availability and IP have been polar opposites for years. Since IP networks have been able to route around problem points, many datacom carriers have avoided meeting traditional system uptime requirements. But this approach has proved costly, forcing some IP network operators to install an entire backup system to ensure network reliability.

"There's a bit of a lost opportunity here," said Jim Guillet, assistant vice president in Alcatel's broadband networks division. "IP acceptance may not be as high as it could be if high availability was added."

The capital and operation expense of adding high availability have hindered the acceptance of IP equipment. To cut down on network development, operation and maintenance costs, carriers must eliminate redundant router architectures, said Hudson Gilner, director of product management at Avici Systems.

The solution for both Avici and Alcatel was to develop nonstop routing software packages that turn traditional IP switch and routing architectures into telco-grade equipment.

Alcatel on Monday (May 20) released its 7670 routing switch platform, its first product featuring the company's ACEIS software. The software, released last month, allows an IP systems to replicate data on two control cards, each with its own database. In this way, a system keeps an exact replica of incoming data so that problems, such as loss of state, do not occur in case of failure.

ACEIS software will ship with new 7670 systems and will be available as a software upgrade for current 7670 customers, Guillet said.

Backup architecture

Avici on Tuesday (May 21) unveiled its iPriori nonstop routing architecture, a backup architecture for the company's SSR and TSR routing platforms. "The route control server is the Achilles heal in routing system designs," Gilner said.

Avici's "fundamentally different approach" implements a fault-tolerant TCP design that enables a backup route server to take over at the point where a route server fails. The backup server contains enough information for TCP connections to be maintained, for a routing table to be rebuilt, and for an MPLS label database to be rebuilt. All software processes are restarted in the backup server, which can operate as a self-contained entity.

A key aspect of this approach is that the backup database does not operate in lockstep with the active route server. In cases where communication systems employ a mirroring approach, active and backup servers run all the same processes, save the same state data, and so on. By operating in lockstep, however, the backup server risks being crippled by the same bug that brings down the active server. Avici avoids this problem by loosely coupling its systems, Gilner said.

Another feature of Avici's high-availability scheme is the ability to perform hitless software upgrades. Designers can upgrade TSR and SSR routers with the nonstop routing scheme without taking the systems down, Gilner said.