Abstract

Microsoft commissioned Tolly to evaluate SNA Server 2.1, with the main focus on benchmarking gateway performance under load against Novell NetWare for SAA 1.3b in large-scale 3270 environments. The project examined how well each gateway maintained mainframe 3270 file-transfer responsiveness as interactive traffic increased, using foreground IND$FILE transfers while simulated background logical units steadily loaded the gateway.  

The December 1994 report states that the evaluation was conducted in three phases: an initial prototype of the test procedure at The Tolly Group lab in Manasquan, New Jersey, a scaled-up round at Microsoft using a beta build of SNA Server 2.1, and a final verification round using a shipping version of SNA Server 2.1 and NetWare for SAA 1.3b with the most recent Novell PTF levels. Tolly emphasizes that no established benchmark standard existed for SNA gateways, so the goal was to measure relative performance under identical conditions rather than produce absolute industry-wide scores. The core workload used a 1MB IND$FILE transfer from the mainframe to a 3270 client, with gateway load increased in increments of five active LUs generated by looping macros on background 3270 sessions.  

Results showed that Microsoft’s gateway maintained much more stable performance as background traffic increased. With no background load, SNA Server transferred the 1MB file in 38.1 seconds versus 39.7 seconds for Novell. At loads up to 10 active LUs, both products behaved consistently. At 15 or more LUs, however, NetWare for SAA performance began to degrade sharply. At 35 background LUs, NetWare for SAA required 199.6 seconds, while SNA Server completed the same transfer in 50.2 seconds, making Microsoft’s result almost four times faster. The chart on page 1 visually reinforces this trend, showing SNA Server remaining near the 40- to 50-second range while Novell’s elapsed time climbs steeply as load grows.  

The test bed used matched Dell 466/ME 486/66 gateways with 32MB of RAM, connected to both 16Mbit/s Token Ring and Ethernet. SNA Server ran on Windows NT Server 3.5 using Olicom Token Ring and Intel EtherExpress adapters; NetWare for SAA ran on NetWare 4.02 with the same hardware. The client station was a Compaq Prolinea 486/50 using Windows for Workgroups 3.11, Attachmate EXTRA! for Windows 4.0, IPX transport, and a 32KB 3270 buffer. The host side used an IBM ES/9000 model 9221-150 connected through an IBM 3745 front-end processor running VM/ESA, VTAM 3.4.0, and NCP 6.2.0. Tolly also notes Microsoft SNA Server 2.1 support for Windows NT Server and Advanced Server, multiple transports including TCP/IP, IPX/SPX, Banyan VINES IP, AppleTalk, RAS, and APPC/CPI-C, plus support for PU 2.0, PU 2.1, APPN LEN node, DSPU, TN3270, and LU 0, 1, 2, 3, and 6.2. Overall, the report presents Microsoft SNA Server 2.1 as the stronger choice for maintaining 3270 responsiveness under heavy gateway load.