Making bandwidth measurements is not entirely self-explanatory, there are many methods and which is best depends both on what type of network is used, and user requirements. Here we present our view on how to make bandwidth measurements in WLAN.
Data rates and efficient bandwidth We would like to define data rate as the rate with which the data are sent through the channel. In WLAN half duplex is used, ie transmit OR receive. At symmetric load the bandwidth that is useful for the user (net bandwidth) is consequently at most half of the data rate. Further more we would like to define efficient bandwidth as the rate the user can use to transmit useful data. Typically, the efficient bandwidth, is less then half of the data rate. IEEE 802.11 BreezeCom FHSS: 3 Mbps data rate gives 1.8 Mbps efficient bandwidth NB! When making measurements on OFDM at 3.5 GHz, we have seen the following phenomena: BreezeACCESS: 12 Mbps data rate gives 4 Mbps efficient bandwidth Wi-LAN Libra 3000: 16 Mbps data rate gives 8-10 Mbps efficient bandwidthMany users have difficulties understanding that data rate is not the same thing as what is shown in the client software. Clarification from the supplier can avoid this problem. Units In general neglect is common when it comes to using units for data rates. kbit/s Mbit/s Mbps and MBps are often used for bit and byte, and sometimes a byte is considered as ten bits (sic!)... Making bandwidth measurements using FTP One method is to send a large file through FTP. The file has to be large to be able to neglect the time for protocol negotiation and so on. We use the client software WS_FTP in Windows 2000 and NcFTP for Linux. As server we exclusively use wu-FTPD for Linux. By using the same test environment we get a reasonable reproducibility for the measurements. The measurement is done clocking a transfer of a large file (100 Mbytes). From this we get the efficient bandwidth (see above) by dividing the time taken for the transfer by the size of the file. A problem we have found, is that when using Windows 2000 we get a lower efficient bandwidth when uploading than we do when downloading. With the same computer, network card, network and so on, we do not get the same phenomena using Linux. This we do not have an explanation for. Making measurements using TPTEST We recommend using TPTEST, that can be downloaded at http://tptest.iis.se. There is a server software (for eg Linux), and also a client software for Windows 2000. This system is established as the de facto standard for making bandwidth measurements in "broadband networks". It establishes a contact between client and server and can test both TCP and UDP. Note that at tests using advanced mode one has to keep track on that the conditions for complying with an approved test. The test must run for at least 10 seconds, and consequently one has to transmit a sufficient amount of data for the test to take at least that time. This method of testing seems to be very stable. Internet is a network of networks. So every ISP only can be held responsible for the part of the network that they run. By our standards it is reasonable to define ones measurements so that one tests the parts of the networks that one has control over. Otherwise it is easy to compare apples to bananas... Asymmetric bandwidth in WLAN The access method for a WLAN (according to IEEE 802.1b) is based on that all stations can hear one another, access point (AP) and all clients. If that is not the case you get the hidden station problem ie all clients can hear the common access point but all clients can not hear each other. Collisions appear and the result is a network with poor performance. In the standard IEEE 802.11b there are functions for RTS/CTS, ie flow control. The client has to make a request to the access point to send (RTS) and wait for a Clear To Send (CTS). Correctly implemented, this method ought to work rather good, but it does not seem to be used at all in a lot of the equipment on the market. The result is that the performance often is good on downloads, ie data traffic from the access point to the client. The maximum efficient bandwidth from an AP according to IEEE 802.11b is typically 4.0-5.5 Mbit/s. At uploads it is usually just around half that bandwidth, over a longer period of time. In reality it is difficult to build a WLAN so that every client can hear the others (and the access method works). What design criteria to use when building surface covering access with WLAN is still an unanswered question, as is what radio make that actually has a functional implementation of RTS/CTS. |
