Each NetCelera pair maintains persistent, logical connections (tunnels) through which packets are exchanged using ITWorx's ACM5 (Adaptive Connection Compression and Multiplexing at Layer 5) architecture. Tunnels are application-specific and dynamically allocated depending on load. ACM5 captures and inspects packets at Layer 5 before they go across WAN links, identifying and eliminating redundant data, such as repeated phrases, blank lines and zeros, to optimize compression ratios. The compressed data is then multiplexed into tunnels to traverse the WAN. On the receiving end, a NetCelera demultiplexes (demuxes) and decompresses the data, and then delivers it.
ITWorx engineers visited our Syracuse University Real-World Labs® and embedded NetCelera appliances on both sides of a simulated WAN link. The WAN united two LANs representing an enterprise headquarters and a remote branch office. A black box (i686) running a Linux 2.4 kernel with 512 MB RAM separated the LANs using Candela Technologies' LANforge ICE WAN Simulator. ICE emulated T1 (1.544-Mbps) and T3 (50-Mbps) links between the HQ and the branch office. It was set to drop and reorder 10 packets out of every 1 million with a 1,024-byte buffer, 60 milliseconds of latency and a random jitter value of 10. This resulted in an average 125-ms delay over the WAN.
In the HQ LAN, I plugged in a Sun Microsystems Sun Fire 280R (dual UltraSPARC III with 750 MHz and 2 GB RAM) running Sun Solaris 9 and an Apache Web server. From there, I mirrored the intranet Web pages for Network Computing. This provided a platform to generate real traffic from the remote-branch LAN using Mercury Interactive Corp.'s LoadRunner 7.5.1 with up to 200 virtual HTTP and FTP users on Dell Intel Celeron (500-MHz) PCs.
Giving It the Squeeze
After running multiple tests, I averaged the results to find a significant bandwidth savings for HTTP data: The NetCeleras reduced an 11-minute test to a 6-minute test, using only 50 MB to send approximately 107 MB of data from the HQ to the remote branch, which equaled a 53 percent savings in bandwidth.