1 / 12

Robert Davis Real-Time Systems Research Group, University of York

Global Fixed Priority Pre-emptive Scheduling: What is the arrival pattern that leads to the worst-case response time?. Robert Davis Real-Time Systems Research Group, University of York. Question scope. Homogeneous Multiprocessor Real-Time Systems Global scheduling Single global run-queue

redford
Télécharger la présentation

Robert Davis Real-Time Systems Research Group, University of York

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Global Fixed Priority Pre-emptive Scheduling:What is the arrival pattern that leads to the worst-case response time? Robert Davis Real-Time Systems Research Group, University of York

  2. Question scope • Homogeneous Multiprocessor Real-Time Systems • Global scheduling • Single global run-queue • Pre-emption and migration • Fixed priority scheduling • Tasks have unique priorities • All jobs of a task have the same fixed priority • What pattern of job arrivals leads to the worst-case response time for a particular task?

  3. Basic task model • Task model • Static set of n tasks tk with priorities 1..n (1 is the highest) • Bounded worst-case execution time Ck • Minimum inter-arrival time or period Tk • Relative deadline Dk • implicit-deadline Dk = Tk • constrained-deadline DkTk • arbitrary-deadline • Independent • Each task gives rise to a potentially infinite sequence of jobs • Worst-case response time Rk is the longest time from arrival to completion of any job of task tk for any valid arrival pattern

  4. System model • Multiprocessor system • m identical processors • Global fixed priority pre-emptive scheduling • At any given time, the m highest priority ready jobs execute • Migration is permitted, but a job can only execute on one processor at a time

  5. Task models • Concrete periodic tasks with synchronous initial release • First job of every task arrives at time 0 • Subsequent jobs arrive strictly periodically Tk after the arrival of the previous job of the same task • Non-Concrete periodic tasks • Arrival times of the first job of each task is unknown • Subsequent jobs arrive strictly periodically Tk after the arrival of the previous job of the same task • Sporadic tasks • Arrival times of all jobs are unknown a priori • Subsequent jobs may arrive at any time once a minimum inter-arrival time Tk has elapsed since the arrival of the previous job of the same task

  6. What is known? • Global FPPS is a completion time predictable algorithm • [Ha and Liu 1994] showed that later completion times (or longer response times) cannot be obtained by reducing the execution time of any job • Worst-case response time for concrete periodic tasks with synchronous /asynchronous initial release • [Cucu and Goossens 2006, 2007] showed that worst-case response times can be determined by simulating the schedule over the LCM of task periods (or longer in the arbitrary deadline / asynchronous case) and using worst-case execution times • However the LCM can be very large…

  7. 1 2 3 4 4 4 What is known? • Synchronous arrival sequence does not necessarily result in the longest response time (unlike in the Uniprocessor case) [Lauzac et al. 1998] • Lower priority tasks have no effect on when higher priority tasks execute • So we can think in terms of when a lower priority task can and cannot run 1 3 P1 2 4 P2

  8. Recent work • Sporadic and non-concrete periodic tasksets (Generalises a more specific result in [Guan et al. 2009]) • Theorem: A worst-case response time for task tk occurs when arrival of a job of task tk at time t is coincident with all m processors becoming busy with higher priority tasks (i.e. in the time interval [t-ε,t)not all m processors were busy) • Proof: (By contradiction) Assume that the worst-case response time occurs for a job of task tk which arrives at time x, not compliant with the theorem, and this response time is strictly greater that that for any such arrival at a compliant time t.

  9. Recent work • Case 1: At time x, not all m processors are busy with higher priority tasks. We can move the arrival time forward to the next compliant time t without decreasing the response time • Case 2: At time x, all m processors are busy with higher priority tasks and have been busy since the last compliant time t. We can move the arrival time back to last compliant time t without decreasing the response time Case 1: Response time Case 2: Response time

  10. Recent work • This theorem tells us something useful about the pattern of arrivals that leads to the worst-case response time • Specific case used to improve sufficient schedulability tests [Guan et al. 2009] • State-of-the-art sufficient tests are still pessimistic with respect to • The amount of higher priority task execution falling within an interval • The time for which that amount of higher priority task execution occupies all m processors

  11. Research Question • What is the arrival pattern of higher priority tasks that leads to the worst-case response time? 1 3 1 1 4 P1 2 5 2 P2 4 6 3 P2 7 Response time

  12. Research Question • Fully determining this arrival pattern provides an exact schedulability test • No exact test is currently known for non-concrete periodic tasksets, except in theory for simulating all distinct combinations of arrival patterns over the LCM, which is intractable for any reasonable sized examples • No exact test is currently known for sporadic tasksets • Any properties of the arrival pattern that we can derive provide an opportunity to improve upon existing sufficient schedulability tests for global FPPS

More Related