and in the area of web hosting, technologies such as load balancing provides the means to
handle huge number of simultaneous requests to services. Although ever increasing
bandwidth and over provisioning can help to improve conditions, there are a number of
reasons why the introduction of QoS is still necessary.
Networks, systems and applications are susceptible to congestion and overload that can
affect data throughput because application traffic is unpredictable by nature. Since these
applications often share the same resources at the same time, congestion is often the
result. Web sites can be hit due to a massive increase in traffic or what is known in the
Internet world as the Slastdot Effect [8]. This often results in what seemed like over
provisioned servers and networks to quickly appear insufficient due to sheer volume of
attempts to reach a site simultaneously. There are simply too many users now on the web
to deliver a first class service all the time based on best effort.
Different applications have different requirements for throughput, reliability, delay and
jitter. Some service are more elastic that others. Although users are normally prepared to
put up with delay with elastic applications
1
because it is expected to be delivered later in
the day and picked up some other time, one may send an urgent email which can be
treated as a real time or inelastic application [9]
.
So it would be more efficient if one
could segment traffic based on its requirements for delay, jitter etc. Even on relatively
unloaded IP networks, delivery delays can vary enough to adversely applications that
have real time constraints [3].
Certain users are willing to pay for a guaranteed service, while others are willing to
suffice with a best effort . Since QoS provides value in the service, it implies the need
for
accounting and billing.
$