Abstract

Akamba Corp. commissioned The Tolly Group to quantify the HTTP (i.e. “Web”) transaction performance and response time gains delivered using its Velobahn High-Performance Web Server Accelerator, a PCI-based accelerator board that integrates Akamba’s MassTrans HTTP acceleration functionality with a standards-based Fast Ethernet network interface controller.The Akamba offering was tested in conjunction with APACHE 1.3.11 Web server running the FreeBSD 3.3 operating system.

This Tolly Group performance evaluation examines the Akamba Velobahn High-Performance Web Server Accelerator, a PCI-based HTTP acceleration board designed to increase Web server capacity by offloading CPU-intensive HTTP processing from the host server. Tested in September 2000 with Apache 1.3.11 on FreeBSD 3.3, the Velobahn integrated Akamba’s MassTrans HTTP acceleration technology with a standards-based Fast Ethernet NIC and was evaluated under simulated e-commerce-style Web traffic representing mixed WAN access speeds.   

Tolly benchmarked the accelerator against the same server equipped only with a standard Fast Ethernet NIC, using simulated loads from 200 to 1,200 concurrent HTTP 1.0 sessions. The traffic mix was modeled to reflect a consumer Web audience: 8% xDSL at 1.5Mbit/s, 42% at 56Kbit/s, 42% at 33.6Kbit/s, and 8% at 14.4Kbit/s. RULER, a FreeBSD-based traffic-generation tool, was used to emulate realistic user sessions and control session rates so that a switched Fast Ethernet lab could mimic varied WAN bandwidth, latency, and loss characteristics. 

The results showed substantial gains in both response time and throughput. With 1,000 simulated users, average response time on the standard server was 18.6 seconds, compared with just 4.4 seconds using Velobahn, a reduction of more than 76%. At 1,200 users, response time improved from 20.9 seconds to 5.8 seconds. Aggregate transaction throughput also rose sharply under heavier loads: at 1,000 users, the standard server processed 46.6 transactions per second, while the Velobahn-equipped server handled 145.3 transactions per second, more than tripling throughput. Across the full test range, Tolly concluded that the accelerator increased Web server capacity by as much as 4:1 while maintaining complete user and network transparency. 

The server under test was a 600MHz Pentium III Network Engines system with 512MB of RAM and Apache 1.3.11, while eight FreeBSD client systems generated traffic through a Cisco Catalyst WS-C2924-XL switch. Overall, the report positions the Akamba Velobahn as a practical way to scale Web infrastructure, improve response times, and reduce the need for additional servers, rack space, and supporting network equipment in busy Web environments.