Predictable Network Computing

Clusters of networked, off-the-shelf workstations are currently used for computationintensive, parallel applications. However, it is hardly possible to predict the timing behaviour of such systems or to give guarantees about execution times. In this paper, we present the design of message-driven scheduling to integrate Quality-of-Service guarantees in the distributed execution of parallel shared-memory programs in the “Shared Objects Netinterconnected Computer (SONiC)” environment. SONiC provides a class-based programming interface; it consists of user-space servers and a runtime library. SONiC’s Scheduling Server allows for predictable partitioning of the CPU cycles of a single workstation. To provide predictable execution in a distributed environment, we extend this approach by a real-time communication server responsible for delivering QoS guarantees to the SONiC tasks by integrating message processing and scheduling. We present measurements which show the impact of the Sched...

• Type: Conference
• Year: 1997
• Where: ICDCS
• Authors: Andreas Polze, Gerhard Fohler, Matthias Werner
• Keywords: Distributed And Parallel Computing, Execution Times, ICDCS 1997, Predictable Execution, SONiC’s Scheduling Server