Synchronization Protocols. Task Allocation Bin-Packing Heuristics: First-Fit Subtasks assigned in arbitrary order To allocate a new subtask T i,j

Similar documents
Embedded Systems. 4. Aperiodic and Periodic Tasks

Real-Time Systems. Multiprocessor scheduling. Multiprocessor scheduling. Multiprocessor scheduling

Last Time. Priority-based scheduling. Schedulable utilization Rate monotonic rule: Keep utilization below 69% Static priorities Dynamic priorities

Preemptive scheduling. Disadvantages of preemptions WCET. Preemption indirect costs 19/10/2018. Cache related preemption delay

Portioned Static-Priority Scheduling on Multiprocessors

Clock-Driven Scheduling (in-depth) Cyclic Schedules: General Structure

Scheduling Motivation

Real-Time Operating Systems M. 11. Real-Time: Periodic Task Scheduling

AN EXTENDIBLE APPROACH FOR ANALYSING FIXED PRIORITY HARD REAL-TIME TASKS

Resource Sharing. CSCE 990: Real-Time Systems. Steve Goddard. Resources & Resource Access Control (Chapter 8 of Liu)

EDF Scheduling for Identical Multiprocessor Systems

Improved Worst-Case Response-Time Calculations by Upper-Bound Conditions

Quantifying the Sub-optimality of Uniprocessor Fixed Priority Non-Pre-emptive Scheduling

Quantifying the Sub-optimality of Uniprocessor Fixed Priority Pre-emptive Scheduling for Sporadic Tasksets with Arbitrary Deadlines

Global EDF Scheduling for Parallel Real-Time Tasks

Partitioned Mixed-Criticality Scheduling on Multiprocessor Platforms

Two Methods to Release a New Real-time Task

Computer Control: Task Synchronisation in Dynamic Priority Scheduling

Overhead-Aware Compositional Analysis of Real-Time Systems

ECE559VV Project Report

Using non-preemptive regions and path modification to improve schedulability of real-time traffic over priority-based NoCs

Offline Equivalence: A Non-Preemptive Scheduling Technique for Resource-Constrained Embedded Real-Time Systems

Solution Methods for Time-indexed MIP Models for Chemical Production Scheduling

Procrastination Scheduling for Fixed-Priority Tasks with Preemption Thresholds

Additional File 1 - Detailed explanation of the expression level CPD

Parallel Real-Time Scheduling of DAGs

An Integrated OR/CP Method for Planning and Scheduling

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

ELG3336: Op Amp-based Active Filters

Energy-Efficient Scheduling Fixed-Priority tasks with Preemption Thresholds on Variable Voltage Processors

Energy-Efficient Thermal-Aware Scheduling for RT Tasks Using T CP N

A Real-Time Multicast Routing Scheme for Multi-Hop Switched Fieldbuses

Improving the Sensitivity of Deadlines with a Specific Asynchronous Scenario for Harmonic Periodic Tasks scheduled by FP

Fixed-Priority Multiprocessor Scheduling with Liu & Layland s Utilization Bound

Hard Real-time Embedded Systems are ubiquitously used today in safety critical commercial applications

Queueing Networks II Network Performance

J. Parallel Distrib. Comput.

Analysis of Queuing Delay in Multimedia Gateway Call Routing

Fixed-Priority Multiprocessor Scheduling with Liu & Layland s Utilization Bound

Keynote: RTNS Getting ones priorities right

Improving the Quality of Control of Periodic Tasks Scheduled by FP with an Asynchronous Approach

Split alignment. Martin C. Frith April 13, 2012

DEADLOCK INDEX ANALYSIS OF MULTI-LEVEL QUEUE SCHEDULING IN OPERATING SYSTEM USING DATA MODEL APPROACH

Minimisation of the Average Response Time in a Cluster of Servers

The Schedulability Region of Two-Level Mixed-Criticality Systems based on EDF-VD

Single-Facility Scheduling over Long Time Horizons by Logic-based Benders Decomposition

Two-Phase Low-Energy N-Modular Redundancy for Hard Real-Time Multi-Core Systems

Planning and Scheduling to Minimize Makespan & Tardiness. John Hooker Carnegie Mellon University September 2006

Distributed Auctions for Task Assignment and Scheduling in Mobile Crowdsensing Systems

Scheduling within Temporal Partitions: Response-time Analysis and Server Design

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud

Lecture 8: Time & Clocks. CDK: Sections TVS: Sections

Limited Preemptive Scheduling for Real-Time Systems: a Survey

Handling Overload (G. Buttazzo, Hard Real-Time Systems, Ch. 9) Causes for Overload

Instantaneous Utilization Based Scheduling Algorithms for Real Time Systems Radhakrishna Naik 1, R.R.Manthalkar 2 Pune University 1, SGGS Nanded 2

Distributed and Stochastic Machine Learning on Big Data

Parametric Utilization Bounds for Fixed-Priority Multiprocessor Scheduling

NP-Completeness : Proofs

Partitioned Scheduling of Multi-Modal Mixed-Criticality Real-Time Systems on Multiprocessor Platforms

Efficient Feasibility Analysis for Real-Time Systems with EDF scheduling*

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

Introduction to Continuous-Time Markov Chains and Queueing Theory

Lecture 4: November 17, Part 1 Single Buffer Management

Problem Set 9 Solutions

Chapter Newton s Method

Hybrid Unicast and Multicast Flow Control: A Linear Optimization Approach

arxiv: v3 [cs.os] 12 Mar 2016

Task Scheduling with Self-Suspensions in Soft Real-Time Multiprocessor Systems

Real-Time Scheduling of Parallel Tasks under a General DAG Model

Partitioned Scheduling of Multi-Modal Mixed- Criticality Real-Time Systems on Multiprocessor Platforms

Critical sections. Using semaphores. Using semaphores. Using semaphores. How long is blocking time? 17/10/2016. Problems caused by mutual exclusion

The strict priority scheduler

O-line Temporary Tasks Assignment. Abstract. In this paper we consider the temporary tasks assignment

Partitioned EDF Scheduling in Multicore systems with Quality of Service constraints

Minimizing Cache Overhead via Loaded Cache Blocks and Preemption Placement

On Energy-Optimal Voltage Scheduling for Fixed-Priority Hard Real-Time Systems

Energy and Feasibility Optimal Global Scheduling Framework on big.little platforms

Dynamic Programming. Preview. Dynamic Programming. Dynamic Programming. Dynamic Programming (Example: Fibonacci Sequence)

Resource Reservation for Mixed Criticality Systems

Lecture 8: S-modular Games and Power Control

Pythagorean triples. Leen Noordzij.

Annexes. EC.1. Cycle-base move illustration. EC.2. Problem Instances

Improvements on Waring s Problem

Analysis of Discrete Time Queues (Section 4.6)

Estimating Delays. Gate Delay Model. Gate Delay. Effort Delay. Computing Logical Effort. Logical Effort

jitter Abstract Output jitter the variation in the inter-completion times of successive jobs of the same

Research Article On the Modelling of the Mobile WiMAX (IEEE e) Uplink Scheduler

Utility Based Optimal Task Scheduling Problem in a Multi-agent System

Toward Understanding Heterogeneity in Computing

Start Point and Trajectory Analysis for the Minimal Time System Design Algorithm

Common loop optimizations. Example to improve locality. Why Dependence Analysis. Data Dependence in Loops. Goal is to find best schedule:

Task Reweighting under Global Scheduling on Multiprocessors

Energy-Aware Fault Tolerance in Fixed-Priority Real-Time Embedded Systems*

A Bayesian Approach to Arrival Rate Forecasting for Inhomogeneous Poisson Processes for Mobile Calls

Calculation of time complexity (3%)

PreDVS: Preemptive Dynamic Voltage Scaling for Real-time Systems using Approximation Scheme

bounds compared to SB and SBB bounds as the former two have an index parameter, while the latter two

Time Synchronization for Wireless Sensor Networks

The optimal delay of the second test is therefore approximately 210 hours earlier than =2.

1 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations

Transcription:

End-to-End Schedulng Framework 1. Tak allocaton: bnd tak to proceor 2. Synchronzaton protocol: enforce precedence contrant 3. Subdeadlne agnment 4. Schedulablty analy Tak Allocaton Bn-Packng eurtc: Frt-Ft Subtak agned n arbtrary order To allocate a new ubtak T,j If T,j can be added to an extng proceor P l (1 l k) wthout exceedng t capacty, allocate T,j to P k Ele add a new proceor P k+1 and allocate T,j to t. Chenyang u CSE 467S 1 Chenyang u CSE 467S 2 Performance lmt of Frt-Ft Number of proceor needed: m/m 0 -> 1.7 a m 0 -> m: number of proceor needed under Frt-Ft m 0 : mnmum number of proceor needed Frt-Ft can alway fnd a feable allocaton on m proceor f total ubtak utlzaton no greater than m(2 1/2-1) = 0.414m Aumng fxed-prorty chedulng, dentcal proceor Tak Allocaton to Mnmze Communcaton Cot Inter-ubtak communcaton can ntroduce overhead and delay E.g., Remote method nvocaton more expenve and lower than local nvocaton Goal: mnmze communcaton cot ubject to proceor capacty contrant Two tep Partton ubtak nto group Allocate group to proceor Chenyang u CSE 467S 3 Chenyang u CSE 467S 4 End-to-End Schedulng Framework 1. Tak allocaton: bnd tak to proceor 2. Synchronzaton protocol: enforce precedence contrant 3. Subdeadlne agnment 4. Schedulablty analy Synchronzaton Protocol Requrement Correct: Enforce precedence contrant Allow accurate chedulablty analy ow wort-cae repone tme ow overhead Reduce jtter ow average repone tme Chenyang u CSE 467S 5 Chenyang u CSE 467S 6

Greedy Protocol Releae job J,j;k a oon a J,j-1;k completed Subtak may not be perodc under a greedy protocol Dffcult for chedulablty analy gher-prorty tak arrve early hgh wortcae repone tme for lower-prorty tak Jtter can accumulate over multple hop Crtque on Greedy Protocol Correctne Allow chedulablty analy Wort-cae repone tme Jtter ow average repone tme Chenyang u CSE 467S 7 Chenyang u CSE 467S 8 Phae-Modfcaton Protocol (PMP) Idea: Enforce perodc releae baed on wort-cae repone tme Every job J,j;k releaed at tme + ( k 1) p j + 1 w l= 1, l : tart tme of job J,1;1 w l : wort cae repone tme of T l Aumpton Requre upper bound on the repone tme of all ubtak Requre global clock Crtque on PMP Incorrect f tak have releae jtter or overrun Allow chedulablty analy ow wort-cae repone tme : No explct ynchronzaton Depend on global clock ynchronzaton ow jtter gh average repone tme Chenyang u CSE 467S 9 Chenyang u CSE 467S 10 Modfed PMP Same a PMP except A ubtak cannot be releaed unle t predeceor ha been completed Aumpton Requre upper bound on the repone tme of all ubtak Requre a ync meage from predeceor ndcatng t predeceor ha been completed Doe not requre global clock ynchronzaton Indcate ahead tme n ync meage Crtque on MPMP Correct Allow accurate chedulablty analy ow wort-cae repone tme requre explct ynchronzaton Doe not requre global clock ync ow jtter gh average repone tme Chenyang u CSE 467S 11 Chenyang u CSE 467S 12

Releae Guard If proceor never dle nce lat releae tme r,k+1;j-1 releae J,k+1;j ether when t receve a ync meage from J,k;j or at tme r,k+1;j-1 +p, whchever later Ele: releae J,k+1;j when recevng a ync meage or when proceor become dle, whchever later. Improve average repone tme wthout affectng chedulablty at the cot of jtter RG Aumpton Aumpton Doe not requre upper bound on the repone tme of all ubtak Doe not requre global clock ynchronzaton Work bet for looely coupled ytem! Chenyang u CSE 467S 13 Chenyang u CSE 467S 14 Crtque on Releae Guard Correct Allow accurate chedulablty analy ow wort-cae repone tme requre explct ynchronzaton Doe not requre global clock ynchronzaton ow jtter (f rule 2 not ued) Improved average repone tme (f rule 2 ued) Chenyang u CSE 467S 15 Score Board: Sync protocol Greedy PMP MPMP RG Correct N Sched N WCRT Avg RT M/ / Chenyang u CSE 467S 16 Global Info. Jtter Ue MPMP or RG f Informaton about all tak are avalable a pror Sytem ha global clock ync Otherwe only RG can be ued Subdeadlne agnment Subdeadlne prorte under EDF & DM repone tme Optmal ubdeadlne agnment NP-hard Offlne: heurtc earch algorthm Onlne: mpler heurtc Subdeadlne Agnment eurtc Notaton (Relatve) deadlne d of tak T (Relatve) ubdeadlne d j of ubtak T j (1 j ) Slack of ubtak T j : j = d j -e j Ultmate Deadlne (UD): d j = d But ome ubtak mut fnh earler than the end-to-end deadlne! Effectve Deadlne (ED): d j = d Agn all lack to 1 t ubtak j 1 d k = 1 k k = j+ e 1 k Chenyang u CSE 467S 17 Chenyang u CSE 467S 18

More eurtc Proportonal Deadlne (PD): ej dj = d e k = 1 k Agn lack proportonally to executon tme Normalzed Proportonal Deadlne d e u( V j, j j = d = ( e ( k 1 ku V, k Agn more lack to ubtak on buer proceor ) )) Schedulng Sngle proceor Perodc tak Fxed prorty v. dynamc prorty Deadlne v. perod Reource contenton Aperodc + perodc tak Dtrbuted ytem Chenyang u CSE 467S 19 Chenyang u CSE 467S 20 Schedulng Aperodc Tak ybrd tak et: perodc tak + aperodc tak Problem: Arrval tme unknown Sporadc tak wth a hard deadlne Inter-arrval tme mut be lower bounded Schedulablty analy: treated a a perodc tak wth perod = mnmum nter-arrval tme Aperodc tak wth a oft deadlne Pobly unbounded nter-arrval tme Goal: mantan hard guarantee on perodc tak Reduce repone tme of aperodc tak Background Schedulng Treat aperodc tak a lowet-prorty tak Advantage Smple Aperodc tak ha no mpact on the chedulablty of perodc tak Dadvantage Aperodc tak have very long repone tme when the utlzaton of perodc tak hgh Acceptable only f Sytem not buy Aperodc tak can tolerate long delay Chenyang u CSE 467S 21 Chenyang u CSE 467S 22 Pollng Server Pollng erver (PS): a perodc tak ued to erve aperodc requet Perod: p Capacty: c Rule Releaed perodcally wth perod p Serve any pendng aperodc requet Supend telf f t ha ued up t capacty, or no aperodc requet pendng Server capacty replenhed to c n the next perod Schedulablty The aperodc requet have the ame mpact on perodc tak a a perodc tak. n tak wth m PS : U p + U U b (n+m) Can have multple PS (wth dfferent perod) for dfferent aperodc requet Dadvantage: If an aperodc requet me the executon of PS, t ha to wat tll the next perod long repone tme. Chenyang u CSE 467S 23 Chenyang u CSE 467S 24

Deferrable Server (DS) Unlke PS, DS preerve unued capacty untl the end of the current perod Better repone to aperodc requet owever, DS mpact on perodc tak dfferent from an perodc tak Utlzaton Bound wth DS U + 2 1/ n Under RMS U b = U + n 1 U + 2 A n : U b = U + ln When U = 0.186, mn U b = 0.652 Sytem chedulable f U p U + 2 ln Chenyang u CSE 467S 25 Chenyang u CSE 467S 26 Ponter Cla hand-out Rate Monotonc A Practtoner' andbook for Real-Tme Analy: Gude to Rate Monotonc Analy for Real-Tme Sytem, Klen et. al. EDF Deadlne Schedulng for Real-Tme Sytem: EDF and Related Algorthm, Stankovc et. al. General ard Real-Tme Computng Sytem, G. Buttazzo. Real-Tme Sytem, Jane u. Chenyang u CSE 467S 27