Abstract
Today's Internet offers (almost) global reachability at low cost.
On the one hand, the "Best-Effort" delivery of packets does
not guarantee any Quality of Service (QoS) level, but it is
often sufficient even for high bit rate transfers. On the other
hand, there is an ever increasing tendency to move value added
services like telephony or video conferencing onto the Internet,
which require bounded packet delay and predictable throughput.
High precision applications like tele-surgery, tele-robotics
or tele-music additionally require extremely low packet loss
rates. Therefore, QoS in terms of short packet delay and low packet
loss will be required in future versions of the Internet, so
called ``next generation networks'' (NGNs), to support these services.
QoS can be achieved by introducing an admission control
(AC) infrastructure into the network. Demanding applications
reserve network resources before transmitting traffic over the network,
at least in high-QoS classes. AC blocks out reservations when
the capacity does not suffice to guarantee the required QoS level
both for the new reservation and for the already
established reservations. Of course, the blocking probability has to be
small in order not to annoy customers such that AC admits all
reservations most of the time. This fact is exploited by the
capacity overprovisioning (CO) approach that simply trades in
complexity cost for bandwidth cost. Instead of blocking flows in
overload situations, the CO approach tries to provide enough bandwidth
so that the resulting QoS violations are low enough to be tolerated
by the relevant applications.
Many investigations compare blocking probabilities and
required capacity for different AC schemes, for which several
signalling protocols exist but none has been introduced on a large
scale on the Internet. Introducing an AC architecture for the Internet
would require significant investment into control plane elements
and operation. Thus, currently all high-quality providers apply
CO, leading to a low utilization of core networks Odly03.
In this paper, we quantify the capacity requirements for
networks relying on AC or CO for QoS provisioning to economically
assess both approaches. We assume the traditional AC approach of
deciding link by link if a reservation can be admitted. We concentrate
on the high priority traffic with given bandwidth requirements per flow
and compute the network capacity required for it. To explain
the fundamental relations between flow bandwidth and link
capacity requirements, we evaluate first the single link scenario.
Then, we extend the study to an entire network domain to reflect
realistic scenarios.
Links and resources
Tags
community