Understanding Wireless Network Performance - Technology Information

Telecommunications, Sept, 2000 by John Fischer

What is the speed on an expressway? The 65-mph speed posted at the roadside, the 5 mph that traffic moves during rush hour, or the 55 mph that cars in the HOV lane experience at that same time?

Even though it is slower than the 65-mph rated speed, the HOY (high-occupancy vehicle) lane has the advantage. While the actual speed is lower, constant flow and fewer vehicles mean that more traffic passes in a given period. The same can be true for high-speed or broadband Internet access. Even though some service providers quote a lower-rated speed, data throughput may exceed that of a competitor with higher advertised speeds.

It is important to understand the key differences between advertised performance and the actual speed delivered. The variance can be due to many factors. The speed discrepancy debate starts to take shape in wireline systems. A 56K dial-up modem does not give the advertised file transfer rate of 7 kbps. Something in the range of 0.2 kbps to 2.0 kbps is more common. T1 and DSL exhibit similar differences between advertised and true performance. T1, for example, claims 1.544 Mbps. The reality is a connection somewhere in the range of 1.2 Mbps to 1.4 Mbps (see Table 1).

The speed of a DSL line depends on the distance from the CO: not only the direct measured distance, but the length of the actual connection route between point A and point B. Other significant factors include the quality of the copper, bundle cross talk, the number of splices in the line, bridge taps and load coils. These factors contribute to a common scenario in which users a few miles away get only a 100-kbps to 200-kbps connection.

With fiber to the desktop, download speed is limited by many external factors including the hardware interface between the PC and the fiber link, the operating system, and application software controlling the transfer. The PC itself, including its CPU and bus speeds, disk access time, and amount of RAM are also factors. The connection between the PC and the link may be a LAN, so factors such as other LAN traffic, congestion, router performance and the speed of the network interface card can produce bottlenecks.

The speed of the server providing the data can also be a major factor. Tests have demonstrated significant differences in performance between servers running Windows and Linux, as well as for different controlling software applications. Moreover, latency induced by the server or client hardware, including disk access time and RAM size, can become the overriding limitation on ultimate link speed.

All these limitations influence dedicated, point-to-point links with no networking overhead. When the link is part of a shared network, multipoint system, networking overhead also comes into play One of the best features of wireless networks is how easily and naturally a shared, multiuser network can be created in the air interface. As with any multiple access network scheme, time delays in the grant of bandwidth to an individual user can and will decrease throughput.

Wireless Discrepancies

Wireless vendors often advertise what is known as "signaling" rate (the rate at which bits are sent over the air interface) as the link speed, which can be misleading. The actual throughput may be significantly less due to overhead from a variety of sources. One source is header information, which consists of addresses, routing information, signal control, forward error correction and QOS bits. Often, this information is added to every packet of data transmitted.

High speeds alone are not enough to create a truly fast link if latency is high. Even in a predominately one-way transmission, such as an FTP (File Transfer Protocol) download, the uplink contains return acknowledgments. The downlink cannot stream more data until the uplink return acknowledgment has been received. Such latency translates into gaps or dead time in the total payload transmission. In a highly asymmetrical link, the slow uplink will serve to limit any FTP download speed. This is true, for example, in a megabit/second" wireless downlink with a telephone line return such as that used in many LMDS or MMDS systems.

Changing the network access scheme, effectively making the payload packets longer so the overhead is a smaller percentage of the total, can reduce overhead. However, the longer a streaming transmission is, the longer the latency is for other users wanting to gain access.

Any network can be optimized for good throughput for streaming applications, but then be proven far less than optimal for thin client-server transactional applications, Because reducing latency is the key element for transactional data transfer, a compromise must be struck. So, when comparing throughput, look closely at the conditions under which performance is measured. Look especially at the assumed packet sizes and number of users on the test network.

Networks designed for metropolitan area applications must account for the speed of light. As swift as it is, the speed of light is slow enough to contribute to overhead. Range delay can produce a hesitation of about 150 milliseconds at a distance of 25 miles. Half-duplex turnaround time is another overhead drain. Most low-cost wireless systems are half-duplex (they cannot transmit and receive simultaneously). The time it takes a system to switch modes adds to the overhead and reduces payload time. When combined with range delay, it can become even more significant.