| Summary: | 碩士 === 國立臺灣大學 === 資訊管理學系 === 86 === Traditional Internet has offered one type of service, i.e. the
best-effort packet delivery. However, many new applications
demand rigid network delivery performance, e.q., timing
constraints. In response, IETF has defined Integrated Services
Network Framework for the next generation Internet. Under it,
two QoS control services are defined: the Guaranteed service
and Controlled-Load service.
In this work, we study two different approaches in the support
of the Controlled-Load service at a network element. The first
approach is called Probabilistic Buffer Preemption (PBP)-based
Congestion Control. When network is under congestion, defined
as buffer overflow, arriving packets with higher priority will
preempt packets of lower priority in the buffer queue according
to some probability distribution. This approach can be viewed a
packet discarding scheme for congestion control. The second
method studied is to take a more rigid resource sharing approach.
In this work, we use Weighted Fair Queueing (WFQ) based packet
scheduling. Performances of these two schemes are studied via
simulation.
The results show that the performance (packet loss probability)
of the Controlled-Load service is mainly determined by the
characteristics and the arrival processes of both the
Controlled-Load and the Best-Effort traffics. Traffic isolation
between the two service classes is poor. When combined with the
use of token-bucket based traffic shaper, the performance of the
scheme is largely improved. Since the WFQ scheduling algorithm
provides excellent fairness of resource sharing, traffic
isolation among different service classes, minimum throughput
guarantee to individual traffic flows and bounded delay, the
simulation results show that this scheme is a good candidate to
support Controlled-Load service. In addition, we study the
queueing behavior of the buffer of the token bucket as well as
the randomization effect and the output process from the WFQ
packet scheduler.
|
|---|
