Title

Application and effects of process residual time information in dynamic time-sensitive scheduling

Date of Completion

January 2005

Keywords

Operations Research|Computer Science

Degree

Ph.D.

Abstract

Time-sensitive scheduling situations abound in computer and networking applications. From classical hard and/or soft real time applications, to the scheduling of Internet traffic in web/database servers, scheduling impacts system and process behavior and remains an important area of research. This research investigates the use of extended job information (e.g., service time or size distribution) in the scheduling problem. If the distribution of job times is known, then the residual time (expected time remaining for a job), rm(t), based on the service it has already received, can be calculated. Often a scheduling decision is based on a single value execution time (e.g., mean or worst case). However, when multiple jobs are concurrently scheduled with time quanta such that their scheduling is based on partial execution, a single value execution time estimate for jobs can be insufficient, especially when time is critical, and may result in undesirable scheduling because a single value representation does not provide an accurate residual time, rm(t), estimate. In prior work task residual time is estimated by simply subtracting the passed time from the initial mean value or worst case without considering the distribution. ^ As an alternative, we have developed a new technique of estimating task residual execution time, rm(t), from the knowledge of the complete execution time and probability distribution. A path dependent rm(t) is updated dynamically using the complete distribution, which can provide a more accurate rm(t) as opposed to using a single value. Our approach provides better estimates of task's residual execution time that can be used by scheduling process at decision making points. ^ In the previous work, there has not been an comprehensive study in: (a) considering a complete execution time distribution; (b) calculating task remaining time accurately from the distribution in a dynamic manner; and (c) observing its effect on scheduling. This model will show the benefits of using distribution properties in scheduling of time sensitive tasks. We developed a Residual Time Based (RTB) scheduling algorithm utilizing the knowledge of job residual time. Extensive simulation studies have been performed to evaluate the performance of RTB in single and dual-processor cases. A broad variety of job service time distributions with a wide range of coefficient of variations and shapes have been considered. We have compared the results with RR (Round Robin), FCFS (First Come First Serve), and EDF (Earliest Deadline First) scheduling approaches and find that in all distributions studied our algorithm is able to satisfy more job deadlines. ^