Topics: Flow and congestion control, routing, scheduling, and buffer management
Authors: Fabian Schwarzkopf, Sebastian Veith and Michael Menth (University of Tuebingen, Germany)
Presenter bio: Michael Menth is professor at the Department of Computer Science at
the University of Tuebingen/Germany and chairholder of
Communication Networks. He received a diploma and a PhD degree in
1998 and 2004 from the University of Wuerzburg/Germany. Previously,
he was studying Computer Science at the University of Texas at
Austin and worked at the University of Ulm/Germany. His special
interests are performance analysis and optimization of
communication networks, resource and congestion management,
resilience and routing issues, industrial networking and Internet
of Things. He holds numerous patents and received various scientific awards for innovative work.
Abstract:
In the recent years, the bufferbloat
phenomenon was observed which is mainly due to oversized unmanaged
buffers in the Internet. This triggered a new discussion of active queue
management (AQM) algorithms in the IETF. "Controlled Delay" (CoDel) and
"Proportional Integral controller Enhanced" (PIE) are considered as an
alternative to "Random Early Detection" (RED). Their intention is both
to take advantage of large buffers for occasional bursts and to limit
queueing delays most of the time. Moreover, they are able to cope with
varying bandwidth. In this paper, we study the performance of CoDel,
PIE, and CoDel-ACT, which is an effective modification of CoDel that
leads to better performance than CoDel in our studies. We experiment
with saturated TCP sources and a fixed-bandwidth bottleneck link and
focus on the delay-limiting phase of the algorithms. We investigate the
impact of configuration parameters and traffic load on link utilization
and queueing delay. We study the timely evolution of queuing delays and
drop patterns, and point out significant differences among the
algorithms. In particular, we show that CoDel's drop behavior changes
over time and may lead to underutilization.
