Network Switch Roundup

The Tests
We engineered the tests to show the true throughput of the switches by avoiding any dependence on other hardware in the test machines. The unloaded test were done by hooking only the two test machines to the switch and running our favorite throughput testing tool, NetCPS. The beauty of NetCPS is that is can be configured to pass any amount of data from point to point without ever using the hard drive. It's small enough to sit completely in memory, and the data it passes is made up on-the-fly rather than in a file sitting on the test machine. We ran three passes of a 500 MB data transfer and three passes of a 30 MB data transfer for each of the switches. The NICs in each machine were Intel 10/100s, and both test machines were Celeron 500s with 128MBs of RAM. The loaded tests were performed with the same two test machines and three machines for creating network traffic. We performed exactly the same data sizes and number of passes as with the unloaded test, of course. To demonstrate the contrast between switches and hubs, we also ran a high quality 10/100 Ethernet hub through the loaded and unloaded 500 MB transfer tests and included the results in the comparison table below. The loading setup is shown in the diagram below:

Here's how it works:

1)	The loading machines, 1 and 2, download a single 1 GB binary file from the FTP Server (red lines)
2)	At the same time, machines 1 and 2 also perform a 1 GB NetCPS transfer between them (green lines) Note: Machines 1 and 2 each have only one physical connection to the switch. The lines are separated to indicate the two separate, silmultaneous processes.
3)	The test machines, A and B, perform the NetCPS transfers recorded in the comparison table below.