Abstract

Multi-vendor report of running SNA protocols over TCP/IP using the Data Link switching approach as outlined in IETF RFC 1434. This test included the following solutions: IBM 6611 Network Processor, Proteon Central Network Exchange (CNS) 600, and Wellfleet Communications' Backbone Concentrator Node (BCN). 

This Tolly Group executive summary presents benchmark results for early RFC 1434-compliant Data Link Switching (DLSw) products, a technology designed to address a critical weakness in traditional token-ring bridging for SNA and LLC2 traffic. In conventional source-route bridged environments, remote PCs access mainframe resources across WAN links using token-ring data-link protocols, but that model offers little protection for session integrity during congestion. Even short periods of WAN overload can cause SNA/LLC2 sessions to fail. DLSw was developed to solve that problem by allowing routers to intervene in previously untouchable protocol exchanges while remaining transparent to end stations. This study evaluated three early implementations: the IBM 6611 Network Processor, Proteon CNX 600, and Wellfleet Backbone Concentrator Node. 

Tolly and Layland Consulting designed the benchmarking to answer two practical questions: how much single-session throughput these DLSw platforms could sustain across a dedicated T1 link, and whether they could preserve 3270 sessions when a 56Kbit/s WAN link was fully saturated. In the T1 performance tests, the lab ran live SNA/APPC file transfers across paired routers connecting two 16Mbit/s token-ring LANs. Throughput was measured using SNA Request/Response Unit sizes of 256, 512, 1,024, and 2,048 bytes. The table on page 2 shows that Proteon led the group, delivering 1.33, 1.39, 1.43, and 1.44Mbit/s respectively, with the final value noted at an RU size limited to 1,920 bytes. IBM measured 0.32, 0.64, 1.26, and 1.41Mbit/s, while Wellfleet posted 0.79, 0.91, 0.91, and 1.29Mbit/s. These results showed that all three platforms approached WAN line-rate performance under favorable conditions, and that Proteon in particular came close to wire speed across all tested RU sizes. 

The report highlights that Proteon’s strong results were aided by “packet grouping,” a technique that buffers smaller frames and loads them into a single WAN packet to improve efficiency. IBM also supported this feature, although the report notes that its implementation did not perform as well with smaller packets, likely because DLSw processing on the 6611 was handled by the main processor rather than a dedicated communications interface processor. Even so, the study concludes that DLSw, despite being more complex than source-route bridging, can still deliver performance across the WAN that is comparable to top-tier token-ring bridges. 

Tolly also tested congestion behavior by establishing a live 3270 session between a mainframe and a PC running Attachmate EXTRA! for Windows 4.0, then saturating a simulated 56Kbit/s WAN link first with IP traffic and then with APPC traffic. Vendor prioritization schemes were deliberately not invoked so the lab could observe raw congestion response. Every product passed this test: no SNA session was lost, even though some sessions paused during the overload interval and resumed when congestion cleared. That result is significant because it shows that DLSw could make mission-critical SNA sessions far more resilient across constrained WAN links. 

The test bed, shown in the diagram on page 3, linked two 16Mbit/s token-ring LANs through a point-to-point WAN emulator operating at 56Kbit/s and T1 speeds. APPC end stations used Hewlett-Packard Vectra 486/33U EISA-bus PCs with Olicom EISA token-ring adapters, while the 3270 station used an HP Vectra 486/33U with a 3Com TokenLink III EISA adapter. The mainframe was an IBM 9370 running VM/ESA release 1.0, VTAM 3.4, and CMS Level 7. Traffic analysis and verification were handled with a Network General Expert Sniffer, a Wandel & Goltermann DA-30, and an HP Series J2300 Protocol Analyzer. Overall, the report presents RFC 1434-compliant DLSw as an important step toward reliable, high-performance transport of SNA traffic across routed wide-area networks.