Abstract

Multi-vendor test report of local Token Ring bridges. Bridges tested from the following vendors: 3Com, Connectware, IBM, Proteon, and SMC.

This Tolly Group executive summary presents the results of its 1994 Industry Benchmark for local token-ring bridges, a category the report describes as a foundational building block of enterprise internetworks because bridges provide the basic mechanism for linking individual LANs into larger LAN/WAN environments. The study was designed to measure how well contemporary source-route bridge products handled the throughput demands of modern 16Mbit/s token-ring networks, and it extended prior testing by also including optional evaluation of IP and IPX routing capabilities. The products tested represented five vendors and included 3Com’s NETBuilder II, Connectware’s TokenSwitch 3000, IBM’s 8229 Bridge, IBM’s LANStreamer Token-Ring Bridge/DOS, Proteon’s ProNET CNX 600, and SMC’s EliteSwitch ES/1. Depending on the platform, testing covered two-port, four-port, and six-port configurations. 

The report emphasizes that results improved dramatically over the previous year’s benchmark. In the 64-byte source-route bridge frame test, every product exceeded the 13,000 frames-per-second threshold, and all but one surpassed 16,000fps. The chart on page 1 shows that the Connectware TokenSwitch 3000 led the group at 23,450fps, followed by IBM’s 8229 Bridge at 21,625fps. SMC’s EliteSwitch ES/1 reached 18,189fps, IBM’s LANStreamer Token-Ring Bridge/DOS delivered 16,700fps, and Proteon’s CNX 600 measured 13,221fps. The report notes that the TokenSwitch 3000’s 23,450fps result also represented the maximum guaranteed frame rate the test bed could produce at that frame size, meaning the bridge was effectively able to forward the full offered load without loss. 

Performance remained strong with minimum-sized 28-byte frames. The page 2 chart shows the TokenSwitch 3000 again in first place at 38,400fps, followed by IBM’s 8229 Bridge at 23,582fps and SMC’s EliteSwitch ES/1 at 23,293fps. IBM’s LANStreamer Token-Ring Bridge/DOS posted 18,686fps, while the Proteon CNX 600 reached 13,673fps. The report highlights that more than half the tested products forwarded 23,000fps or better at this size, with all but one handling at least 18,000fps. These results are significant because 28-byte and 64-byte frames stress forwarding engines far more than larger packets and often expose limitations in bridge architecture more clearly than Mbps-oriented measurements alone. 

At larger frame sizes, the products moved even closer to wire-speed operation. The page 3 chart, which summarizes 256-byte two-port source-route bridge throughput in Mbit/s, shows Connectware at 14.84Mbit/s, IBM LANStreamer at 14.58Mbit/s, IBM 8229 Bridge at 14.57Mbit/s, SMC EliteSwitch ES/1 at 13.74Mbit/s, and Proteon CNX 600 at 12.29Mbit/s. Tolly concludes that with frame sizes of 512 bytes and above, almost every product delivered 90% of available wire speed or better. The report also notes that four-port and six-port test results were similarly encouraging: all products evaluated in those multi-port configurations maintained their two-port performance for frame sizes of 256, 512, 1,024, 2,048, and 4,096 bytes. That suggests that backplane capacity and forwarding logic were robust enough to preserve throughput even as port count increased. 

Tolly identifies Connectware’s two-port TokenSwitch 3000 as the overall top performer in the benchmark. According to the summary, it forwarded 38,400 28-byte frames and 23,450 64-byte source-route bridge frames without packet loss, and it demonstrated near-wire-speed performance, which the lab defined as 90% or more of bandwidth, when handling 256-byte frames and larger. The report states that the TokenSwitch 3000 was able to forward all frame rates the test bed could generate. IBM’s products also performed particularly well at larger frame sizes. Both the IBM 8229 Bridge and IBM LANStreamer showed wire-speed behavior for 256-byte frames and above, finishing second and third respectively in that class. SMC’s EliteSwitch ES/1 is described as posting solid performance throughout the program, ranking well at smaller frame sizes and taking top honors in all multi-port testing among the products evaluated there. Proteon’s CNX 600 participated in every test and delivered 80% of wire speed or better when handling 512-byte frames and larger. Tolly also reports an early result for 3Com’s NETBuilder II, using development code planned for 1995 release, showing 14,184fps for 64-byte frames, about four times faster than 3Com’s result in the prior year’s Industry Benchmark. 

The methodology section explains that the test bed was built around two, four, and six 16Mbit/s token-ring segments, depending on whether two-port, four-port, or six-port testing was being performed. The diagram on page 4 illustrates the six-port setup, with multiple source rings feeding the bridge under test and corresponding destination rings connected to analysis tools. The rings were connected by Multi-station Access Units, and PC-based frame generators on the source rings produced valid source-route bridge traffic at frame sizes of 28, 64, 256, 512, 1,024, 2,048, and 4,096 bytes. IP and IPX frames were generated at 64, 256, 512, 1,024, 2,048, and 4,096 bytes. Four- and six-port tests focused on the 256-byte-and-larger range. Throughput was determined by transmitting a known frame rate on the source rings and counting output frames per second on the destination rings. The maximum throughput without frame loss was defined as the point at which source-ring frame rate differed from destination-ring frame rate, with a 0.05% error allowance to account for measurement uncertainty. Instrumentation included a Wandel & Goltermann Domino DA-320 WAN analyzer for frame counts, plus a Network General Sniffer and Siemens K-1100 analyzer for frame-size and fps verification. 

Overall, this benchmark presents a strong picture of progress in local token-ring bridge performance during 1994. Compared with earlier years, the tested products showed marked gains at small frame sizes while also sustaining near-wire-speed forwarding at larger sizes and across multi-port configurations. For network managers building or expanding token-ring-based enterprise networks, the report suggests that several bridge platforms had reached a level of maturity capable of supporting demanding internetwork traffic without the performance compromises that had characterized earlier generations.