Session 9: Distributed Operating Systems

Abstract

Tlie aim of this session is to present some very interesting experiences related to various aspects of tlie impleiiieiitatioii of operating systems in parallel aiid distributed processing eiiviroiimeiits. Tlie iiiiportance of the runtime support in these eiiviroiuneiits is well understood: being tlie software layer between tlie liardivare architecture aiid the application software, it is up to the operating sj.steii1 to (try to) solve all problems related to Iiard\\iare independence, workload distribution, inter-processor coiiuiiunication, perfoniiance optimization, real-time behavior, etc. The first paper of tlie session - Sorne Ixszies ,for the Distributed Scliedziling Problem in tlie A402 Distributed Rea2-Tinie Object-0rierired ,Yj~tern. b!) B. Mecibah and A. Attoui (France) - concentrates on tlie scheduling problem of tlie cictive objects available in tlie MO2 object-oriented real-time model. aiming at supporting implemeiitatioli of a rcal-time database iiianageiiieiit systeiii. Active objects are characterized by an autonomous: event-driven behavior, iiidepeiideiit of the activation of their methods. Tlie proposed solution is a distributed scheduling algorithm adopting a heuristic approach for optimization. Tlie second paper of the session - A /.)is/r/bii/ed Algori tlim ,for E’nzi I/- 7 blercinr Ilynnm ic 7 i7.r k S’chedziling, by A. Baucli. E. Maehle and F.J. Mal-k~is (Geniiaiiy) - considers again the problcm of scheduling, but the aim is in this case to obtain a fault tolerant behavior in a ~iarallel s!.stcm \\.ittiout requiring a static redundancy based on full replication of tasks. Tlie basic idea for acliieving this fault tolerant behavior is to keep all input data sets of a