Summary: Probabilistic End­to­End Delay Bounds for
Earliest Deadline First Scheduling
Matthew Andrews
Bell Laboratories
600­700 Mountain Avenue
Murray Hill, NJ 07974
andrews@research.bell­labs.com
Abstract---We analyze the Earliest­Deadline­First (EDF) scheduling dis­
cipline within the framework of statistical multiplexing. We derive tech­
niques for bounding the probability of delay violations when the session
injections are independent. This enables us to determine whether a given
set of sessions can all meet their delay bounds with the required violation
probability. These techniques can be used by a Connection Admission Con­
trol (CAC) scheme to decide whether to admit a new session. Our analysis
applies to both the single node problem and the network problem in which
the sessions have multiple hops.
We also give extensive numerical results to illustrate how our bounds
may be calculated and to compare the results with estimates that have been
derived for GPS. In addition we show that by altering the deadlines for EDF
we can match the desired violation probabilities more closely.

Source: Andrews, Matthew - Mathematics of Networks and Systems, Mathematical Sciences Research Center, Bell Laboratories

Collections: Mathematics; Computer Technologies and Information Sciences