Flexibility Driven Scheduling and Mapping for Distributed Real-Time - PowerPoint PPT Presentation

Flexibility Driven Scheduling and Mapping for Distributed Real-Time Systems Paul Pop, Petru Eles, Zebo Peng Department of Computer and Information S cience Linköpings universitet, Sweden 1 /23 1 of 14

Outline � Introduction � Incremental design process � Mapping and scheduling � Problem formulation � Mapping strategy � Experimental results � Conclusions 2 /23 2 of 14

Introduction � Characteristics: � Incremental design process, engineering change; � Distributed real-time embedded systems; Heterogeneous architectures; � Fixed priority pre-emptive scheduling for processes; static cyclic scheduling for messages; � Communications using a time-division multiple-access (TDMA) scheme: H. Kopetz, G. Grünsteidl. TTP-A Protocol for Fault-Tolerant Real-Time S ystems. IEEE Computer ‘ 94. � Contributions: � Mapping and scheduling considered inside an incremental design process; � Two design criteria (and their metrics) that drive our mapping strategies to solutions supporting an incremental design process; � Two mapping algorithms. � Message: � Engineering change can be successfully addressed at system level. 3 /23 3 of 14

“Classic” Mapping and Scheduling I/ O Interface RAM ROM CPU AS IC Comm. Controller S S S S S S S S 0 1 2 3 0 1 2 3 S lot TDMA Round Cycle of two rounds 4 /23 4 of 14

Incremental Design Process � S tart from an already existing system with applications: � In practice, very uncommon to start from scratch. � Implement new functionality on this system (increment): � As few as possible modifications of the existing applications, to reduce design and testing time; � Plan for the next increment : It should be easy to add functionality in the future. 5 /23 5 of 14

Mapping and Scheduling Do not exist yet Future at Version N! applications Map and schedule so that the Current Version N+1 future applications applications will have a chance to fit. N No modifications are performed to Existing N-1 t he exist ing applications applications. 6 /23 6 of 14

Mapping and Scheduling Example Existing applications Current apps Future apps Processor P 1 P 3 P 2 P 4 m 1 m 4 Bus m 2 m 5 m 3 7 /23 7 of 14

Mapping and Scheduling Example, Cont. Existing applications Current apps Future apps P 3 P 2 Processor P 1 P 4 Bus m 5 m 4 m 1 m 2 m 3 8 /23 8 of 14

Mapping and Scheduling Example, Cont. Existing applications Current apps Future apps P 3 P 4 P 1 Processor P 2 Bus m 1 m 2 m 4 m 3 m 5 9 /23 9 of 14

Problem Formulation Input � A set of exist ing applications modelled as process sets. � A current application to be mapped. � Each process in the application has its own period, priorit y and deadline . � Each process has a pot ent ial set of nodes to be mapped to and a WCET . � The system architecture is given. Output � A mapping and scheduling of the current application , so that: � Requirement a: constraints of the current application are satisfied and minimal modifications are performed to the exist ing applications. � Requirement b: new fut ure applications can be mapped on the resulted system. 10 /23 10 of 14

Mapping and Scheduling, Requirement a) � Mapping and scheduling of the current application , so that: Constraints of the current application are satisfied and minimal modifications are performed to the exist ing applications. � Subset selection problem elect that subset Ω of existing applications which guarantees that the S current application fits and the modification cost R( Ω ) is minimized: ∑ Ω = R ( ) R i Γ ∈ Ω i 11 /23 11 of 14

Characterizing Existing Applications Γ 1 Γ 2 Γ 4 Γ 5 Γ 1 Γ 2 Γ 4 Γ 5 70 50 150 70 Γ 3 Γ 6 Γ 3 Γ 6 50 Γ 8 Γ 9 Γ 10 Γ 7 Γ 7 Γ 8 Γ 9 Γ 10 20 R({ Γ 7 })=20, R({ Γ 3 })=50, R({ Γ 3 , Γ 7 })=70, R({ Γ 4 , Γ 7 })=90 (the inclusion of Γ 4 triggers the inclusion of Γ 7 ), R({ Γ 2 , Γ 3 })=120, R({ Γ 3 , Γ 4 , Γ 7 })=140, R({ Γ 1 })=150, .... The total number of possible subsets is 16. 12 /23 12 of 14

Mapping and Scheduling, Requirement b) � Mapping and scheduling of the current application , so that: New fut ure applications can be mapped on the resulted system. P 4 m 5 � Design criteria reflect t he degree t o which a design meet s t he requirement b); � Design metrics quantify the degree to which the criteria are met; � Heuristics to improve the design metrics. 13 /23 13 of 14

Characterizing Future Applications 100 100 90 90 ] ] Probability [% Probability [% 80 80 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 0.02 0.05 0.1 0.2 2 4 6 8 Typical ut ilizat ion fact ors Typical message sizes [bytes] U f =C/ T � S mallest expected period T min � Expected necessary bandwidth b need 14 /23 14 of 14

Mapping and Scheduling: Processes � Design criterion for processes: available ut ilizat ion � How well t he available ut ilizat ion of t he current design alternative accommodate a family of f ut ure applications that are characterized as outlined before; � Design metrics for the first design criterion P for processes � C 1 � How much of the largest f ut ure application (total available utilization), cannot be mapped on the current design alternative; � Bin-packing algorit hm using t he best -f it policy : utilization factors of processes as obj ects to be packed, and the slack as cont ainers. P =40% C 1 15 /23 15 of 14

Mapping and Scheduling: Messages � First design criterion for messages: slack sizes � How well t he slack sizes of t he current design alternative accommodate a family of f ut ure applications that are characterized as outlined before; � Tries to cluster t he available slack: t he best slack would be a cont iguous slack. � Design metrics for the first design criterion m for messages; � C 1 � How much of the largest f ut ure application (contiguous slack), cannot be mapped on the current design alternative; � Bin-packing algorit hm using t he best -f it policy : processes as obj ects to be packed, and the slack as containers. contiguous slack m =0% C 1 a) m =0% C 1 b) m =75% C 1 c) 16 /23 16 of 14

Mapping and Scheduling: Messages, Cont. � S econd design criterion: slack distribution for messages � Used for the reduction of design space exploration � How well the slack of the current design alternative is distributed in time to accommodate the messages of a family of fut ure applications; � Tries to dist ribut e the slack so that we periodically ( T min ) have enough necessary bandwith b need for the most demanding future application. � Design metrics for the second design criterion m is the sum of minimum periodic slack inside a T min period on each processor. � C 2 T min b need m =0 < b need =40ms a) C 2 m =40 ms b) C 2 17 /23 17 of 14

Mapping and Scheduling Strategy � Initial mapping and scheduling � Requirement a) Minimizing the modification cost R( Ω ) , subset selection : � Exhaustive Search (ES) � Ad-Hoc Solut ion (AH) � Subset Selection Heuristic (SH) � Requirement b) Starting from a valid solution, heuristics to minimize the obj ective function: = + + − P P m m m m C w ( C ) w ( C ) w max( 0 , b C ) 1 1 1 1 2 need 2 � Ad-Hoc approach (AH), little support for incremental design. � Simulated Annealing (SA), near optimal value for C. � Mapping Heuristic (MH): � Iteratively performs design transformations that improve the design; � Examines only transformations with the highest pot ent ial to improve the design; � Design transformations: moving a process to a different processor, moving a message to a different slack on the bus. 18 /23 18 of 14

Experimental Results How do the subset selection algorithms compare? Average Modification Cost R( Ω min ) 1200 AS 1000 S H 800 ES 600 400 200 0 160 240 320 400 480 Number of processes in the current application exist ing applications: 400 19 /23 19 of 14

Experimental Results How does the quality (cost function) of the mapping heuristic (MH) compare to the ad-hoc approach (AM) and the simulated annealing (S A)? Deviation from near optimal 160 AM 140 MH 120 S A 100 80 60 40 20 Avg. % 0 40 80 160 240 320 Number of processes in the current application exist ing applications: 400 20 /23 20 of 14

Experimental Results, Cont. Are the mapping strategies proposed facilitating the implementation of future applications? No modifications allowed of fut ure applications mapped 100 MH* Future 80 application AM 60 Current application 40 20 Existing application 0 % 40 80 160 240 Number of processes in the current application exist ing applications: 400, future application: 80 21 /23 21 of 14

Experimental Results, Cont. Are the mapping strategies proposed facilitating the implementation of future applications? Modifications allowed of fut ure applications mapped 100 AM Future MH* 80 application MH 60 Current application 40 20 Existing application 0 % 40 80 160 240 Number of processes in the current application exist ing applications: 400, future application: 80 22 /23 22 of 14