Modelling a master detail scheduler for the laboratory

Transcription

1 Fachhochschule Wiesbaden Department 06 Computer Science Modelling a master detail scheduler for the laboratory Reinhold Schäfer 1

2 Agenda Scenario and definitions Scheduling and re-scheduling Master detail scheduling Our testbed Conclusions 2003 Reinhold Schäfer 2

3 Fachhochschule Wiesbaden Department 06 Computer Science Scenario and definitions QC environment 3

4 Scenario: QC environment Many products Different workflows per product Workflow steps atomic One step uses one instrument at one time Workflow steps need access to the same devices 2003 Reinhold Schäfer 4

5 Scenario: definitions (1) Workflow: Several steps blonging to the same test Robot: open hand Robot: move to sample Robot: grasp sample Robot: move to balance Robot: ungrasp sample Balance: weigh sample Robot: grasp sample... Atomic step (activity): One instrument at one time 2003 Reinhold Schäfer 5

6 Scenario: definitions (2) Workflow execution Activities in a sequence Activities in a directed graph Different colors = different instruments 2003 Reinhold Schäfer 6

7 Scenario: definitions (3) Activity constraints Timing constraints End-to-start end-to-end start-to-start Conditional constraints Activity can be executed only if one or several conditions are true 2003 Reinhold Schäfer 7

8 Scenario: definitions (4) Instrument: analytical device with one or more functions Example: balance Open door Close door Tare Weigh Transfer result 2003 Reinhold Schäfer 8

9 Scenario: QC environment A B C D instruments green sample orange sample t gr1 time violett sample 4711 t v1 Conflict, if t gr1 = t v1 or overlap 2003 Reinhold Schäfer 9

10 Fachhochschule Wiesbaden Department 06 Computer Science Scheduling / re-scheduling 10

11 Definition: scheduling Process of generation of resource specific execution plans based on a given set of activities and constraints 2003 Reinhold Schäfer 11

12 Process environment (1) Overlapping resource utilization Robot Balance Preparation Execution Post-execution time 2003 Reinhold Schäfer 12

13 Process environment (2) Overlapping resource utilization Time for different sample Activity delay constraint R. B. t < tmax time 2003 Reinhold Schäfer 13

14 Process environment (3) Overlapping resource utilization Activity delay constraint Non-interruptible sequence Robot Pipettor Incubation Photometer t < t NIS 2003 Reinhold Schäfer 14

15 Process environment (4) Overlapping resource utilization Activity delay constraint Non-interruptible sequence Pooling Accumulation of samples running on a multi-position instrument Sample A, 10 min Sample C, 30 min Sample B, 20 min 2003 Reinhold Schäfer 15

16 Process environment (5) Overlapping resource utilization T = 76 C Sensor Activity delay constraint Non-interruptible sequence Pooling yes T > 80 C no Control events during runtime 2003 Reinhold Schäfer 16

17 Process environment (6) Overlapping resource utilization Activity delay constraint Non-interruptible sequence Pooling Control events during runtime Collision detection and recovery Robot Robot 2003 Reinhold Schäfer 17

18 Algorithmic approaches FIFO scheduling Run scheduler every 10 sec Real-time resource assignment (first come first serve) scheduler does not find best solution in most cases Optimizing scheduling Brute force calculation of all permutations of activities Select best alternative need for pruning heuristics to reduce search trees 2003 Reinhold Schäfer 18

19 Run-time events User e.g. ad-hoc samples R B G T Execution plans R B G T Work bench e.g. failure of a device 2003 Reinhold Schäfer 19

20 Definition: re-scheduling Process of generation of resource specific execution plans based on a given set of activities and constraints taking actual activities into account 2003 Reinhold Schäfer 20

21 Re-scheduling problems Calculation time depends on Number of different workflows Number of instruments Number of optimization goals 2003 Reinhold Schäfer 21

22 Re-scheduling FAQ (1) What happens if an instrument fails? Put all affected samples into parking positions. What happens to samples running on non-affected instruments? Continue with execution. What happens to samples scheduled to run on an unavailable resource? Execute them until last executable NIS and park it Reinhold Schäfer 22

23 Re-scheduling FAQ (2) What happens if a resource fails in the middle of a NIS? Throw sample away Reinhold Schäfer 23

24 Re-scheduling vs. recovery Scheduling algorithms may need too much time for calculation. Find a local solution: recovery. (IF spare time is available THEN backtrack to solvable state) Example: Balance s door doesn t open try it again Reinhold Schäfer 24

25 Master detail scheduling: problem description Instr. A Instr. B Prepare and test micro titer plate Master plan (work bench) Detail plan (workstation) 2003 Reinhold Schäfer 25

26 Master detail scheduling: today s solution In case of problems in detail schedule: Plan for sufficient spare time in master schedule Or Skip sample Consequences: Waste of time not really relevant for long workflows Need for manual recovery 2003 Reinhold Schäfer 26

27 But... Large number of samples to be procecced Processing times are becoming much shorter ratio between processing time and spare time in unacceptable ranges Only small amounts of sample material available need to recover failing steps Consequences: Need to automate master detail scheduling Necessity to propagate the recovery between detail and master level 2003 Reinhold Schäfer 27

28 Master detail scheduling: recovery Instr. A Instr. B Prepare and test micro titer plate Master plan (work bench) Detail plan (workstation) Set up rules to solve erroneous states locally Propagate failure to master plan if recovery fails 2003 Reinhold Schäfer 28

29 Master detail scheduling: pro s and con s Detail plan influences master plan vice versa Small number of activities Local calculation efficient Optimal schedule via brute force calculation 2003 Reinhold Schäfer 29

30 Fachhochschule Wiesbaden Department 06 Computer Science Scheduling testbed with model trains 30

31 Modelling the testbed Lab scheduler calculates optimal sequence of activities (one plan per instrument) samples instruments Model Train = sample Coach = activity to be executed on a device Model train scheduler determines correct route 2003 Reinhold Schäfer 31

32 Scheduling for model trains 2003 Reinhold Schäfer 32

33 Model train testbed results (1) Multi-level scheduling with the laboratory and railway scheduler Generation of optimal activity sequence by lab scheduler Generation of optimal train route by railway scheduler Goal oriented switching of freight cars via generateand-test method Search tree with heuristical pruning 2003 Reinhold Schäfer 33

34 Model train testbed results (2) Shadow monitoring Cooperation between simulation and real world Real-time process control 2003 Reinhold Schäfer 34

35 Conclusions High complexity of scheduling processes Need to include various laboratory requirements Challenges in using master detail paradigm for the reduction of calculation complexity 2003 Reinhold Schäfer 35

36 Acknowledgements Andreas Besier Martin Peschke Christian Piotrowski Andreas Necknig Alexander Rühl Thorsten Richter Bernd Stiehl Stefan Tömö Agilent Technologies Germany GmbH Aventis Pharma GmbH Beckman Coulter, Inc. Gerhardt GmbH Hewlett Packard Germany GmbH TECAN AG 2003 Reinhold Schäfer 36

37 Fachhochschule Wiesbaden Department 06 Computer Science Questions? 37