This paper describes how profile-driven data compression, a very effective approach to reduce memory and bus traffic in singletask
embedded systems, can be extended to the case of systems offering multi-function services.
Application-specific profiling is replaced by static data characterization, which allows to cover a larger spectrum of the
system’s input space; characterization is performed by either averaging several profiling runs over different application
mixes, or by resorting to statistical techniques. Results concerning memory traffic show reductions ranging from 10% to 22%,
depending on the adopted data characterization technique.
This work was supported in part by HP Italiana S.p.A. under grant n. 398/2000.