Abstract

IBM commissioned The Tolly Group, as part of its 1994 Industry Benchmark for PCMCIA LAN performance testing, to evaluate the IBM PCMCIA Credit Card Adapter II Ethernet with the main focus on measuring notebook Ethernet performance versus a desktop ISA-adapter baseline in both ODI and NDIS environments. The project examined how closely a mobile PCMCIA Ethernet adapter could approach desktop-class throughput under Novell NetWare and IBM OS/2 LAN Server conditions, using a one-to-one client/server test model over 10Mbit/s Ethernet.  

The November 1994 Technology Spotlight explains that Tolly’s benchmark was part of its first comprehensive performance study of Ethernet and Token Ring PCMCIA LAN adapters. The report notes that mobile users had many product choices but little performance data, and that the goal was to determine whether notebook users gave up meaningful LAN performance compared with traditional desktop systems. Tolly compared the IBM PCMCIA Ethernet adapter against a baseline established with ISA-bus adapters in desktop PCs and tested the card with both Open Data-link Interface and Network Driver Interface Specification drivers.  

In the ODI suite, the IBM adapter was tested at 64, 256, 512, 1,024, and 1,512-byte frame sizes with packet burst disabled, plus a maximum-frame packet-burst test. The chart on page 1 shows IBM throughput of 0.8, 1.7, 2.6, 3.6, and 5.4Mbit/s respectively, versus baseline results of 1.0, 2.3, 3.5, 5.0, and 6.5Mbit/s. With packet burst enabled at the 1,512-byte maximum frame size, the IBM adapter reached 9.5Mbit/s versus a 9.6Mbit/s baseline, essentially matching desktop performance. Tolly states that at the maximum frame size with packet burst enabled, the adapter was within 0.91% of baseline, while at other tested frame sizes it ran within about 17% to 27% of baseline performance.  

In the NDIS suite shown on page 2, the IBM PCMCIA adapter again tracked closely to baseline. IBM measured 0.8, 1.0, 1.5, 2.3, and 8.7Mbit/s at 81, 256, 512, 1,024, and 1,514-byte frames respectively, compared with baseline results of 0.9, 1.2, 1.8, 2.8, and 8.8Mbit/s. Tolly says the adapter was within 1.37% of the baseline at maximum frame size with packet burst and within about 12% to 16% at other sizes. The test bed used a Toshiba T1910CS notebook with 12MB RAM, PhoenixCARD Plus 3.00, NetWare 3.12 or IBM DOS LAN Requester 3.0, and a 10Mbit/s Ethernet LAN instrumented with a Network General Expert Sniffer and Hewlett-Packard Series J2300 analyzer. Overall, the report presents the IBM PCMCIA Credit Card Adapter II Ethernet as a strong-performing notebook LAN solution that came close to desktop Ethernet throughput, especially at larger frame sizes.