I am doing 802.11b performance testing. Does TCP throughput depend upon file size? In my tests, throughput for 20 KB files was about 0.5 Mbps; throughput for 800 KB files was about 4 Mbps, tested under the same conditions.

    Requires Free Membership to View

    Login

Performance testing can be very tricky. As you've noticed, measurements must be taken over a long enough period to obtain "steady state" results. When you transfer a very short file, the time required to establish the FTP control session, then launch the FTP data session, represents a very significant portion of the total transfer duration. During this FTP client/server back and forth, the wireless link is going unused, so you are not measuring the link's maximum throughput. When you transfer a large file, more/most of the test is spent actually sending large packets of data, "filling the pipe" between the FTP client and server. Thus, your 4 Mbps result more closely represents the actual capacity of the link than your 0.5 Mbps result.

There are many other factors that influence FTP throughput, including maximum segment size (MSS), TCP window size, the content of the file itself, the number of hops between client and server, utilization of links between those hops, utilization of the FTP client and server systems, and NIC configuration. You should eliminate all other running applications on the client and server system, make sure that your FTP client and server are the only systems using the wireless and wired network, minimize the number of hops between then (including hubs and cable runs).

Use large enough files so that each FTP session runs for minutes, not seconds. Repeat your tests using identical files, dropping high and low results, then averaging the rest, until results appear to be consistent. Testing in this fashion can yield directly-comparable measurements -- for example, to compare throughput measured by a given station at 30 feet vs. 60 feet, or to compare throughput for the same station using two different wireless NICs. The fewer variables you change between tests, the more directly comparable your results.

This was first published in June 2005

Join the conversationComment

Share
Comments

    Results

    Contribute to the conversation

    All fields are required. Comments will appear at the bottom of the article.
    Back to top