Abstract

This is the full technical/engineering report for this project. This report was delivered to the vendor participants. It was only released by The Tolly Group to the general public in 2025 because of possible value to researchers.

For decades, corporations have chosen private leased line networks to link remote sites to corporate headquarters. While this approach usually delivered low bandwidth at high cost, in most cases it was the only available solution. Recently, many organizations have begun migrating to public frame relay for both economic and technical reasons: First, frame relay networks usually cost less than leased line networks. Second, frame relay permits users to add bandwidth dynamically for a fraction of the cost of dedicated bandwidth.

Of course, if Fortune 1,000 network managers are to embrace frame relay, they must be certain that it can support their mission-critical SNA sessions.

Until now, there has been very little empirical data to show how SNA behaves over frame relay. With this in mind, The Tolly Group launched the industry’s first multi- vendor benchmark specifically designed to measure SNA performance over a frame relay network. The tests measured SNA data throughput, 3270 response time, and session integrity over a 56 Kbit/s WAN.

The benchmark participants included leading vendors of routers and Frame Relay Access Devices (FRADs): ACC, IBM, Motorola, Netlink Inc., Presticom (ACT Networks), and Sync Research.

Test Results: Summary
Overall, products demonstrated a high level of SNA performance, stability, and consistency. Some products transported SNA using Data Link Switching, some used RFC 1490 encapsulation, and still others relied upon bridging with local acknowledgment. However, regardless of the SNA transport technique, throughput remained consistently above 80% for all products at all SNA frame sizes, with some products exceeding 90% utilization. Response times remained below two seconds, even during periods of heavy congestion with other SNA or IP traffic. Furthermore, even after 10 repetitions of the test, response times varied by less than one second. It is noteworthy that all products were configured to use a single DLCI and PVC.