Latency

Previous

1.3. Latency

Suppose that you have to go to the bank because you need some money. After having arrived cashiers were clear and you have the job done just in 6 minutes.

Two or three days later you have the same problem, but when you arrive to the bank cashiers are congested (too many people need money too) and queues are long enough. This time you have to wait for your turn and you get your money after waiting 35 minutes.

The example above could be used to explain what latency is. When networks are not congested, this means, transmission capacity of the network is greater than flow of data that has to be moved, latency is very low. It’s like when you reach your bank and cashiers are clear.

But when networks get congested, and it ocurrs frequently, data has to wait and make its queues to be served (to be transported from your workstation to the place where you are interested it goes to), and latency can be very high, provoking problems.

Latency is measured in seconds, or milliseconds (ms), because with networks things go fast. But not very fast at all. If network are not congested latency could be 50 ms. This means that information you send from your workstation to some place takes 50 ms to reach it.

But when networks are congested because everybody want to send and receive data, and capacity of transmitting is less than required, latency could be 300, 400, 1000 ms, or even, several seconds. When this ocurr some applications are really in big troubles, especially those called "playback" applications.

As its name imply a playback application is intended for playing back what is ocurring at sender, as similar as possible at the receiver. Voice is one of these kind of application that is influenced a lot by latency. Imagine you are trying to talk with a friend, and due to technical problems, his/her voice (and yours too) is delayed for 2 or 3 seconds. You talk, and nobody answer, and you could think that something is going bad and repeat again your sentences. But, when doing this, you listen your friend answer of something that you said some time before. Conversation begins to become complicated because it's not easy to follow it.

Something similar occur with multimedia application. If your children are playing some of those war games like "counter strike" by Internet, having a latency of 2, 3 or more seconds, it will destroy the interactivity required to play the game well. These high latencies could ocurr due to the "playback latency effect". For example, MPEG, the common video compression standard, requires what is called an I-frame to decode a GOP (group of pictures). GOP is composed of a sequence of IBBPBB frames. The first frame of the sequence (an I-frame), is so big (7000 bytes) that expand through 7 or more packets. The GOP can have 8000 or more bytes and all of them must reach the destination for decoding to take place. Latency in this case will correspond to the GOP size, and will be higher if we compare it with latency of a single packet alone.

Some applications are less vulnerable to latency. If you are downloading a big file from the Internet perhaps latency doesn’t matter. But not all applications are as permissives, and especially multimedia applications (voice and video) are very restricted in their capacity to support latency.

Another problem is what is called a timeout. If some application is waiting for some information, and this information is not available after some time (timeout time), the information will be unusable at all and having it after the timeout is useless. Receiving application losses the information it requires on time and its usefulness is lost too.

Devices that transport your packets of information through the Internet have to enqueue them, this mean, they have to put them in queues before dispatching. This is a necessary approach to deal with congestion. When some burst of packets reach a router and it can’t dispatch them because lines are congested, it has two choices: delete the packets (dropping) and the information will be lost, or it can try to enqueue the packets, waiting a little to see if lines alleviate and it can send them after a while. The formation of these queues in routers and switches is the main reason why the latency phenomenom appears.

Previous

Content

Next