Abstract

The Tolly Group conducted this 1994 Industry Benchmark study with the main focus on measuring the performance of PCMCIA Ethernet and Token Ring LAN adapters against ISA-based desktop baseline adapters in both ODI and NDIS environments. The project was designed to determine whether notebook users sacrificed meaningful LAN throughput when moving from desktop systems to credit-card-sized PCMCIA adapters, and to provide comparative data across multiple vendors and protocols.  

The November 1994 report describes what Tolly calls the first comprehensive performance study of Ethernet and Token Ring PCMCIA LAN adapters. The products tested included 3Com’s PCMCIA EtherLink III 3C589B, IBM’s PCMCIA Credit Card Adapter II Ethernet and Token Ring, Kingston Technology’s EtherX PCMCIA KNE-PCM, Madge Networks’ Smart 16/4 PCMCIA Ringnode, National Semiconductor’s InfoMover NE4100 Ethernet, and Thomas-Conrad’s TC5141 PCMCIA Ethernet and TC4141 PCMCIA Token Ring. Tolly measured the adapters in one-to-one client/server tests using Novell NetWare ODI drivers and IBM OS/2 LAN Server Advanced 3.0 NDIS drivers, with ISA-bus adapters used as the comparison baseline.  

Overall, the report concludes that PCMCIA adapters performed nearly as well as their ISA counterparts. On page 2, Tolly states that in most cases the PCMCIA products processed data within 20% of baseline, depending on frame size and whether packet burst was enabled, and that in several cases credit-card-sized adapters performed within 1% of baseline. The Ethernet ODI chart on page 1 shows especially strong results at larger frame sizes and with packet burst enabled. Tolly notes that National Semiconductor’s InfoMover NE4100 was the fastest Ethernet ODI product at maximum frame size with packet burst, followed closely by IBM’s PCMCIA Credit Card Adapter II and Thomas-Conrad’s TC5141. When packet burst was disabled, 3Com’s 3C589B led across all tested frame sizes and even exceeded the ISA baseline at the 1,512-byte frame size.  

For Token Ring ODI, page 3 shows IBM and Thomas-Conrad sharing the lead ahead of Madge, except at the 4,096-byte frame size with packet burst disabled, where Madge was strongest. In the NDIS tests, Tolly says the vendor ranking remained generally unchanged, though throughput was slightly lower than in the ODI environment. The Ethernet NDIS chart on page 4 and Token Ring NDIS chart on page 5 both show the leading PCMCIA adapters clustering closely below the desktop baseline, with the gap narrowing at larger frame sizes.  

The test methodology used a Toshiba T1910CS notebook with 12MB of RAM as the PCMCIA client, a Compaq Deskpro 486/66 server with 8MB of RAM, and separate Ethernet and 16Mbit/s Token Ring test beds. Network General Expert Sniffer and Hewlett-Packard Series J2300 analyzers were used to capture traffic and verify frame sizes. ODI Ethernet testing used frame sizes from 64 to 1,512 bytes, while ODI Token Ring testing ranged from 64 to 4,096 bytes. NDIS Ethernet used frame sizes from 81 to 1,514 bytes, and NDIS Token Ring used 81 to 4,200 bytes. Overall, the executive summary presents PCMCIA LAN adapters as viable, high-performing solutions for mobile users, with performance that in many cases closely approached desktop ISA adapter levels.  

Solutions tested:

• 3Com PCMCIA EtherLink III 3C589B — PCMCIA Ethernet adapter tested in the benchmark.  
• IBM PCMCIA Credit Card Adapter II Ethernet — PCMCIA Ethernet adapter included in the comparative tests.  
• IBM PCMCIA Credit Card Adapter II Token Ring — PCMCIA Token Ring adapter tested in the benchmark.  
• Kingston EtherX PCMCIA KNE-PCM — PCMCIA Ethernet adapter included in the evaluation.  
• Madge Smart 16/4 PCMCIA Ringnode — PCMCIA Token Ring adapter tested in the benchmark.  
• National Semiconductor InfoMover NE4100 Ethernet — PCMCIA Ethernet adapter included in the comparative tests.  
• Thomas-Conrad TC5141 PCMCIA Ethernet — PCMCIA Ethernet adapter tested in the benchmark.  
• Thomas-Conrad TC4141 PCMCIA Token Ring — PCMCIA Token Ring adapter included in the evaluation.