As long as Cisco Systems leads the router market, competitors will continue to come up with differentiators. With Adtran's new NetVanta 4305 access router, the differentiators are price, price, price.
Depending on configuration, the NetVanta 4305 -- which Adtran will ship in Australia in Q1, 2005 -- can cost thousands of dollars less than similarly equipped Cisco models.
As our exclusive Clear Choice tests show, the NetVanta 4305 is a modest performer. The beta routing software still has some rough spots, including scalability issues, but the price advantage might offset these, especially for managers of small or midsize enterprise networks.
The 1U systems we tested were equipped with two fixed-port Fast Ethernet interfaces and an optional module supporting eight T-1 (1.544M bit/sec) serial interfaces. The NetVanta 4305's base price includes support for IPv4 routing protocols and a stateful firewall. IPSec support is optional, as is a DSX-1 module for interconnection with PBXs.
The base NetVanta 4305 costs US$2,500, and the system as tested is priced at US$5,000. In contrast, a comparable Cisco 3725 base model lists at US$8,500.
The NetVanta's command-line interface strongly resembles IOS, but lacks some features of its Cisco counterpart, such as redirecting command output through a pipe.
We measured performance of the NetVanta 4305 in seven ways: static routing; small- and large-table routing information protocol (RIP) routing; small-, medium- and large-table open shortest path first (OSPF) routing. We also tested IPSec tunnel capacity.
The static routing test was a best-case scenario; the goal was to show the maximum rate at which the NetVanta 4305 boxes would forward traffic without dynamic routing enabled. We then repeated the same test using either RIP or OSPF, and routing tables of various sizes. We tested in a back-to-back configuration, linking two routers with up to eight T-1 interfaces, and then repeated the tests on one router with traffic flowing between two Fast Ethernet interfaces.
In the two-router tests, throughput was slightly lower than line rate with medium or large frames, regardless of the presence or absence of dynamic routing. Tests with 256-byte frames are the most noteworthy, because that size is close to the average frame length on many enterprise networks.
These tests uncovered two anomalies. First, Adtran's beta software supported only seven T-1s in the multilink PPP link, when dynamic routing was enabled.
Second, throughput with RIP routing didn't scale as high as OSPF, especially when we threw short frames at the NetVanta pair. Short-frame throughput with a 240-route table (the largest we attempted) was noticeably lower than with a smaller 64-route table.
Even with RIP's 15-hop limit, our "maximum" test case represented a relatively small network. Route redistribution and multiple paths easily can swell table size well beyond the maximum levels we used with the NetVanta 4305. With OSPF, throughput for both small and large frames degraded as table size grew.
Degraded throughput with RIP also was pronounced when we offered traffic between two Fast Ethernet interfaces on one router, where frame rates are much higher. However, tests with short frames place the greatest strain on the device being tested, and no production network carries traffic made up exclusively of 64-byte frames.
Then again, even an entry-level US$400 PC can saturate a 100M bit/sec circuit. Given the relatively low cost of processing power, control-plane routing events should have little if any effect on data-plane packet forwarding. Moreover, the choice of routing protocol should not have a marked effect on throughput.
Adtran officials say the company is working to optimize the routing code in the final release of this new code at the end of this month.
Another key metric -- latency -- improved in most cases when we enabled OSPF or RIP. For example, in two-router tests with static routing, we measured average latency of 2.017 millisec when forwarding 64-byte frames. But with RIP routing and 240 routes, average latency actually fell to 1.352 millisec.
In general, latency tests of a single router moving packets between Ethernet interfaces showed very constant results: about 0.460 millisec for 256-byte frames for most test cases. However, when we used OSPF and a 5,000-entry routing table, latency shot up nearly threefold to 1.439 millisec. Latency in the low milliseconds is unlikely to affect performance of any application by itself, but latency is cumulative. In a network made up of many NetVanta routers, latency might grow with routing table size.
The NetVanta 4305 doesn't match Cisco 3700 series routers when it comes to features or robustness of its routing code, but then again it doesn't cost nearly as much. While the routing code we tested had a few unresolved issues, the NetVanta 4305 might be a cost-effective alternative in small to midsize networks.
Newman is president of Network Test, an independent benchmarking consultancy.