-
[show abstract]
[hide abstract]
ABSTRACT: Recent research in real-time divisible load theory (RT-DLT) has addressed the problem of distributing arbitrarily parallelizable real-time workloads among processors which become available at different instants. In prior work [1], we proposed a Linear Programming based formulation of the problem of computing the completion time of a real-time workload on a given collection of processors. In this paper, we show via extensive experimental evaluation that this LP based formulation significantly improves on the heuristic approximations [8,9] that were the only techniques previously known for solving these problems.
Information Technology, 2008. ITSim 2008. International Symposium on; 09/2008
-
[show abstract]
[hide abstract]
ABSTRACT: Recent research in real-time divisible load theory (RT-DLT) has addressed the problem of distributing arbitrarily parallelizable real-time workloads among processors which become available at different instants in the future. Given a real-time job and the times as which the processors become available, we devise exact efficient algorithms to solve two important problems: (i) determine the smallest number of processors needed to complete this job by its deadline; and (ii) given a specific number of processors, determine the earliest completion time for the job on these processors.
Embedded and Real-Time Computing Systems and Applications, 2008. RTCSA '08. 14th IEEE International Conference on; 09/2008
-
Proceedings of the ISCA 20th International Conference on Parallel and Distributed Computing Systems, September 24-26, 2007, Las Vegas, Nevada, USA; 01/2007
-
[show abstract]
[hide abstract]
ABSTRACT: Current real-time application systems demand complex functionality and are increasingly coming to be implemented upon multiprocessor platforms, with complex synchronization, data-sharing and parallelism requirements. However, the formal models for representing real-time workloads have traditionally been designed in the context of uniprocessor environments; hence, they are often not able to accurately represent relevant features of multiprocessor real-time systems. Researchers have recently addressed this shortcoming by applying workload models from Divisible Load Theory (DLT) to real-time systems. The resulting theory, referred to as Real-time Divisible Load Theory (RT-DLT), holds great promise for modeling an emergent class of massively parallel real-time workloads. However, the theory needs strong formal foundations before it can be widely used for the design and analysis of real-time systems. In this paper, we briefly describe our current findings on RT-DLT and ongoing research efforts at extending this work to develop such formal foundations.
-
[show abstract]
[hide abstract]
ABSTRACT: This paper presents a new deadline-based algorithm for dynamic task scheduling on multiprocessor real-time systems. The proposed algorithm was built based on the well-known dynamic scheduling algorithm the Earliest Deadline First (EDF). We extend the EDF algorithm in three significant criterions: (1) The original EDF deals with set of periodic task. (2) It schedules only independent tasks. (3) It works best in the single processor system. Our contribution in this study is a creation of a new variation of EDF called EDF-pc that able to schedule set of non-periodic tasks. It also deals with tasks that may or may not have precedence constraints and it produced acceptable results in the multiprocessor systems.
