As email travels from sender to recipient, it passes thru one or more
mail servers on the network.  At MIT, the servers (also known as hubs)
process the mail and deliver it within a few minutes to the next hop
to its final destination. The average delay for about 400,000
deliveries per day is approximately 8 minutes. If the recipient is
another MIT user, this destination is usually another type of mail
server called a post office. The mail is delivered here within seconds
and stored until picked up by the recipient. However a message can be
queued on a server and not immediately delivered for several reasons.

A server may become overloaded and need to store messages in a queue
until it can catch up later when the load goes down.  Overloading can
occur when someone sends a mass mailing, or "spam". Another cause is
when a host goes "out of control" by sending a message repeatedly in a
rapid-fire manner.  This has been known to happen with so-called
"vacation" programs for example. . here isn't much which can be done 
to guard against such barrages.  When these situations occur, the
servers will queue mail and possibly go as far as refusing to accept
new mail in order to protect the system.

Sometimes a site cannot be reached. This is a routine occurrence
somewhere on the Internet and the message destined for this site will
be queued for later delivery. If a significant portion of the Internet
is unavailable, thousands mail servers outside of MIT will queue mail
destined for MIT. On December 4 this amount of mail was so
extraodinary that the MIT post offices had a difficult time keeping up
with messags sent to MIT users. The result was a few thousand messages
queued for several hours.

The important part of this to keep in mind is that while we do have an
email system here at MIT that usually delivers mail very rapidly,
there is NOTHING in the technology used that guarentees this response
time.  Message queueing and the delays that result from this, are a
normal response of the system to overloads or other
pertubations. Network Operations does its best to avoid and minimize
the impact of these occurrences but users need to set their
expectations accordingly.