Understanding QoS 7
so there has to be some method by which they decide who gets those seats. The gate agent
could swing open the door to the plane and tell everyone to rush onto the plane; whoever
gets to the first class seats first gets them. Some would argue that would be the most fair
way to handle the seating, but that would be very disorderly and probably not a very
pleasant thing to watch. Anyone who flies regularly knows that frequent flier miles are
valuable because you can earn free flights and so on. If you collect enough frequent flier
miles from a specific airline in a single year, however, you will be moved into an elite
frequent flier status and get some extra benefits. One of those benefits is typically some
method by which the most frequent fliers are able to upgrade their coach seat to a first class
seat, when available. Imagine paying full price for a coach ticket to Hawaii, and having no
chance at all to upgrade, while the person beside you is able to upgrade just because he is
a frequent flier. Some would argue that this is unfair. However, it is unfair in a very
controlled way, because there is a specific policy in place that dictates who is eligible for
this upgrade and who is not. This is managed unfairness.
In QoS, managed unfairness is important because sometimes it is necessary to allocate
more bandwidth to one application than another. This doesn’t specifically indicate that
either application is more or less important than the other; rather it indicates a different level
of service that will be provided to each application. That is, the applications may well have
different bandwidth needs, and dividing available bandwidth equally between the two
applications, although fair, might not produce the best results. A good example of such a
scenario is the case of an FTP flow sharing a link with a VoIP flow. The FTP flow is charac-
terized by a large bandwidth requirement but has a high tolerance for delay, jitter, and
packet loss; the VoIP flow is characterized by a small bandwidth requirement and has a low
tolerance to delay, jitter, and packet loss. In this case, the FTP flow needs a larger share of
the bandwidth, and the voice flow needs bounded delay and jitter. It is possible to provide
each flow with what it needs without significantly impacting the service provided to the
other flow. In this case, the allocation of bandwidth is unfair, because the FTP flow will get
more bandwidth, but it is unfair in a very controlled manner. Again, this is an example of
the need for managed unfairness.
Predictability: The Goal of QoS
The successful management of bandwidth, delay, jitter, and packet loss allows for the
differentiated treatment of packets as they move through the network. Unless an implemen-
tation error occurs, all implementations of the differentiated services architecture should
provide the same treatment to each packet of the same type when those packets pass
through a given interface. In Figure 1-1, multiple packets are sent from Bob to the web
server, marked with IP precedence 2.