Cisco A100, FORE ASX 100, SynOptics Lattis Cell Switches Tested

SAN JOSE, Calif., Sept. 12, 1994 -- A series of tests conducted earlierthis month by LANQuest Labs found significant differences in theperformance of three vendors' ATM switches.

Results from LANQuest's tests of the Cisco Systems A100, the FORESystems ASX 100 and the SynOptics Lattis Cell 10114-SM showed that systemthroughput can be significantly degraded by cell congestion and resultingpacket loss through an ATM switch. The Cisco A100 showed the bestsustained throughput, with zero cell loss at loads of 125 Mbps and below.

The three switches were tested to determine cell congestion and loss inboth bursty traffic (characteristic of Local Area Networks) and uniformlydistributed data traffic environments. Switches were tested at two IPframe sizes, 1500 and 4470 bytes, representing the typical maximum framesize used by Ethernet and Token Ring LAN software, respectively. The testconfiguration consisted of four ATM source ports converging to one ATM sinkport. In an application environment this would mirror a client/serverconfiguration. Among specific findings:

• While the Cisco A100 handled both bursty and uniformly distributed traffic well, the FORE ASX 100 showed a significant decrease in sustained throughput for bursty traffic.

• At a source rate of 125 Mbps, the Cisco A100 showed no cell loss, the FORE ASX 100 exhibited degradation under bursty traffic, and the SynOptics Lattis Cell showed the greatest degradation among the three switches.

Robert Buchanan, LANQuest's senior vice president of businessdevelopment, said, "While ATM is moving rapidly into the networkingmainstream, a remaining barrier to its widespread adoption is a lack ofperformance and interoperability guidelines to help users make realisticproduct choices. Seemingly subtle design differences can have profoundeffects on a switch's ability to transmit different data types effectively.These test results take the first step in establishing such guidelines.

"For example, unlike voice and video traffic, which is fairlypredictable, data traffic generated by workstations or routers is verybursty and thus more susceptible to loss. Repeated loss and retransmissioncan lead to significant throughput degradation. Users deploying ATM indata-intensive applications will need to understand that handling burstytraffic isn't just a matter of physical buffer size, but of sophisticatedmechanisms that can increase buffering effectiveness beyond what isphysically available at each output port. In the absence of congestioncontrol standards, which the ATM Forum isn't expected to finalize untilmid-1995, effective buffering levels will be the best measure of a switch'sability to transport real-world, bursty data traffic."

LANQuest develops and markets network test management tools andservices for client-server and internetworking systems. LANQuest alsoincludes LANQuest Labs, an independent network testing facilityspecializing in network components, systems and applications testing.Primary customers include network product vendors and Fortune 2000companies. LANQuest was founded in 1986 and is privately held. Majorinvestors are MK Global Ventures and Aspen Venture Partners.