The knowledge represented by a traffic matrix is very valuable
to a wide variety of traffic engineering tasks including load balancing,
routing protocols configuration, dimensioning, provisioning
and failover strategies. Information on the size and locality of
flows is crucial for planning network growth and diagnosing problems.
Because traffic matrices are not available to carriers today,
they cannot quantify the cost of providing QoS as opposed to overprovisioning.
Building such network-wide views is central to be
able to efficiently engineer an IP backbone network.