WAN latency and what to do about it

You can't control latency.  The minimum latency is based on how far the remote site is away from you, the number of hops, and to a point the maximum speed of the link between the two points.  There is a bit more to it than that, but lets start simple.

The maximum latency is based based on the minimum latency plus the wait/queue time.

Say you work at a fast food restaurant and think of latency is the amount of time it takes you to service a customer.  The timer starts when the customer gets in line.

Now, no matter what you do, there is a minimum amount of time that it takes you to service the customer.  Lets say that is 5 seconds.  So if there are no customers in line and somebody walks up to you, it will take you 5 seconds to service them.  That is you base, minimum, latency.

Now, lets say that all of a sudden there are 10 customers that come at once.  Will the 1st guy will get 5 seconds, but the second guy will get 10, the 3 guy will get 15, and so on until the last guy and he get 55 seconds.

Now to make it more complex, lets say that the service time is also based on what the customer orders (it take less time to transmit a packet that is 100 bytes than it does to transmit a packet of 1500 bytes).  So customer #1 order the simplest item and it take 5 seconds, but customer #2 orders something that take a bit longer, so it takes you 8 seconds. Well customer #2 latency is now 13 seconds (#1 5 + #2 8).

Basically, the more utilized your link is the higher the latency will go.  Either you reduce the volume of traffic, or  increase the amount of available bandwidth.

I think I have it and this makes a kind of sense. Can latency be affected by taks being performed by the client? For example, on a terminal server connection, a user printing a large document off the server to a local printer...would this temporarily increase WAN latency at their location? Enough to make a ping return jump from 86ms to 300ms?

We do not own the network at both ends. :( Connections are run through T-1 at the server end to (typically) through 3 Mb Down / 512 K Up ADSL at the client ends (maximum transmission rate) We are trying to cut down on the periodic lags in the remote desktop sessions. Could I have some suggestions as to what to do about this?

I am not at all familiar with MPLS! I will research it here in a moment.

I can affect latency.  If the traffic has to flow over the WAN link, then yes, it will increase the volume of traffic going over the WAN, which increases bandwidth utilization.  If it increases it to the point where traffic is queued up to go across the WAN, then latency increases, because latency includes queue (waiting in line) time.

MPLS is great for prioritizing traffic, but normally MPLS based networks are more WAN connection that appear to be private site to site connections.  Not a normal Internet type connection, which it appears you have.

You would also have to be able to identify the traffic: print from interactive.  IIRC when you print via term serv, the traffic is sent from the term server to the term client and then forwarded to the printer.  In that case the print traffic looks just like the interactive traffic and MPLS (or any other type of QOS setup) can't tell the difference between the interactive and print traffic.