There are three components to latency that you should combine to come up with the total delay:

1. Propagation delay owing to distance - The amount of time it takes to travel from point A to point B via speed of light. You can calculate this with .01 msec/mile (estimate of speed of light in fiber). This value becomes significant on cross country long haul circuits or international circuits. You need to double your distance if you are calculating roundtrip delay, e.g., NY to SF is approx. 3000 miles or 30ms one way, 60ms roundtrip.

2. Transmission delay - This is the amount of time it takes to move a packet of data across a fixed speed link. To calculate this you would need to estimate or measure an average size packet that shall be transversing the frame relay net. Convert packet bytes to bits and divide by the access speed of the frame relay net to give you seconds. Multiply this result by two since you need to access the frame relay net and come out the other side. If you are calculating roundtrip multiply this number by 4.

3. Queueing delay - This is usually the most significant component of latency. It is also nearly impossible to calculate easily. This component depends on queue sizes in both the frame relay switches and routers, the amount of traffic queued, the number of frame relay switches to be transversed, etc. The number of switches that you transverse will vary based on the condition of the carrier's network at any given point in time. The component doesn't exist on point-to-point serial links built on TDM technology. To estimate queueing delay, do extended ping tests at various times during the day and come up with an average. Routers have built in queuing delay.

As for bandwidth and latency: set up a sniffer to monitor the traffic to and from a workstation then use that station to ping (50+ times) the near and far ends of a WAN link with various packet sizes, around 60, 100, 500, 750, 1000, 1250, and 1500. You need to try at least two different sizes, and the larger the difference the better. Have the sniffer display delta time and packet size and ensure that each ping request is followed immediately by a ping response. Print to a CSV file and import into Excel. Sort the data and analyze everything but the ping responses.

You can use the frequency function to generate a histogram of the response times for each set of pings. The near-end pings will probably be close together but the far-end pings will vary. The distribution will vary depending on the backbone technology and the number of intermediate hops/NNIs in their network. You should have a few pings grouped together at the beginning of the distribution and use them to represent the so-called 'ideal' ping time with no queueing delay. If you don't see this, you may have to increase the number of pings or perform them at a different time of day. You can compute both bandwidth and latency based on the 'ideal' ping times, but that's a different write-up. Basically, it's easier to base things on 'ideal' times.

Rotational latency is the time for a platter in a hard disk to rotate in order to place the desired sector under the read/write heads. Average latency is slightly more time than it takes for half of a platter to pass under the heads.

La"ten*cy (?), n. [See Latent.]

The state or quality of being latent.

To simplify the discussion, I shall distinguish three degrees of this latency. Sir W. Hamilton.


© Webster 1913.

Log in or register to write something here or to contact authors.