Topics: Performance, queuing and scaling models for networks; Performance analysis, bounds and optimization; Traffic characterization and traffic models for networks
Authors: Florin Ciucu (University of Warwick, United Kingdom); Felix Poloczek (University of Warwick / TU Berlin, Germany); Jens Schmitt (University of Kaiserslautern, Germany)
Presenter bio:
Jens Schmitt is a professor in the Computer Science department at
the University of Kaiserslautern, Germany. He is the head of the
Distributed Computer Systems Lab (DISCO). His research interest are in
performance analysis and evaluation of wired and wireless networked and
distributed systems, with an emphasis on quality of service guarantees,
as well as in security for wireless networks.
Schmitt received his diploma degree in CS from the University of
Mannheim, Germany (1996), as well as a Master's degree in Management
Sciences from the University of Swansea, UK (1994). In 2000 he received
his PhD from Darmstadt University of Technology, Germany for his thesis
on "Heterogeneous Network QoS Systems".
Abstract: Promising perspectives of a hypothetical `Tactile Internet', or
`Internet at the speed of light', whereby network latencies become
imperceptible to users, have (again) triggered a broad interest to
understand and mitigate Internet latencies. In this paper we revisit the
queueing analysis of the versatile Markov Fluid traffic model, which
was mainly investigated in the 1980-90s, yet
with questionable accuracy. We derive upper bounds on the tail
distribution of the queue size, which improve state-of-the-art results
by an exponential factor $O(\kappa^n)$, where $0<\kappa<1$ and $n$
is the number of multiplexed sources; additionally, we provide the
first lower bounds. The underlying results are quite general in that
they can be easily adapted to derive the delay distribution for SP,
FIFO, and EDF scheduling. Our overall results rely on a powerful
martingale methodology which was recently shown to be remarkably
accurate.
