Abstract

3Com commissioned Tolly to evaluate the NETBuilder II, with the main focus on measuring large-scale Data Link Switching performance for transporting SNA over IP internetworks under heavy load. The project examined whether headquarters-site routers from 3Com, Cisco Systems, and IBM could maintain reliable throughput and acceptable session response time while supporting live session traffic from more than 250 SNA controllers running up to 504 logical units simultaneously.

The October 1995 report describes this as the most comprehensive multi-vendor DLSw scalability benchmark Tolly had conducted to that point. Tolly tested 20 different configurations, varying the number of physical units and the size of SNA request/response units. The benchmark used IBM’s Transaction Processing Network Simulator to emulate SNA sessions and measured both data throughput and terminal response time. Each configuration was run at traffic levels targeting 25%, 50%, 75%, and 100% utilization of the WAN links. A baseline was also established by connecting all controllers to a single Token Ring without a router or WAN link in order to measure the maximum performance of the TPNS software itself.

The report shows that 3Com’s NETBuilder II forwarded traffic at 75.00%, 86.33%, 95.64%, and 89.70% of the T1 baseline for request/response unit sizes of 256, 512, 1,024, and 1,920 bytes respectively. In the 56Kbit/s low-speed link tests, the NETBuilder II achieved throughput approaching line speed for all data sizes except 256-byte units, posting 49.59%, 74.86%, 78.38%, and 77.30% of the baseline. Tolly also measured response consistency. The baseline 90th percentile and average response times were 0.18 seconds and 0.05 seconds, while the 3Com unit delivered average 90th percentile and mean response times of 1.79 seconds and 1.37 seconds.

The test bed used three branch office routers connected to a headquarters router over three WAN links, with DLSw transporting SNA over the WAN and compression disabled for all products. Each branch router connected to its own 16Mbit/s Token Ring, while an IBM mainframe at headquarters connected through two 3172 controllers and branch-office Token Ring connections through 3745 front-end processors. Tolly also introduced IP congestion to measure the effect of competing traffic on response time and found that the NETBuilder II’s response times increased by 33.65% under those conditions. Overall, the report presents the 3Com NETBuilder II as a strong large-scale DLSw platform with high WAN utilization efficiency and measured response-time behavior under both normal and congested network conditions.

Key technical details in the report include software code level 8.1; support for Ethernet, Fast Ethernet, Token Ring, FDDI, PPP, Frame Relay, ISDN, and X.25 LAN/WAN topologies; routed protocol support including TCP/IP, Novell IPX, DLSw, LLC2 and NetBIOS, XNS, OSI, DECnet Phase IV and V, Banyan VINES, APPN, and AppleTalk Phase II; and additional features such as SDLC conversion, data compression, load sharing, data prioritization, LLC2 IP tunneling, and FRAD support for IBM 3745 FEP connectivity.  

Solutions tested:

• 3Com NETBuilder II — DLSw router evaluated for large-scale SNA-over-IP performance under load.  
• Cisco Systems router — Competing headquarters-site DLSw router included in the large-scale benchmark.  
• IBM router — Competing headquarters-site DLSw router tested for comparative SNA throughput and response time.