24
Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Embed Size (px)

DESCRIPTION

Decomposition using Passages RuleDescription R1Incident arcs of places are equivalent R2Incident arcs of invisible transitions are equivalent R3Incident arcs of visible transitions with non-unique labels are equivalent to incident arcs of all transitions with the same label PAGE 2

Citation preview

Page 1: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Decomposing Replay Problems:A Case Study

Eric Verbeek and Wil van der Aalst

Page 2: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Replay

In• Event log

• Traces− Events

− Activity• Petri net

• Labeled transitions− Activity

• Initial marking• Final markings

• Cost structure

Out• Alignments

• Transition sequence− Activity

− Synchronous− Net only

− Silent transition• Trace

− Activity− Synchronous− Log only

• Minimal costs

PAGE 2

Page 3: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Decomposition using Passages

Rule DescriptionR1 Incident arcs of places are equivalent

R2 Incident arcs of invisible transitions are equivalent

R3 Incident arcs of visible transitions with non-unique labels are equivalent to incident arcs of all transitions with the same label

PAGE 3

Page 4: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Decomposed Replay

• Decompose Petri net into subnets• Using passages (subsuming activities)• Adapt cost structure

• Decompose Event log into sublogs• Empty traces!

• Replay every sublog on corresponding subnet• Idea: Replaying sublogs on subnets is more efficient

than replaying event log on Petri net• Can easily be done on multiple cores

• Merge replay results• Aggregate costs

− SumPAGE 4

Page 5: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Case Study – Settings – Petri nets

Petri net Transitions Places Arcs LabelsRepairExample 12 12 26 8

A32 32 32 74 32

Bpic2012A 11 14 28 10

Bpic2012 58 44 124 36

PAGE 5

Page 6: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Case Study – Settings – Event logs

Event log Cases Events LabelsRepairExample 1104 11,855 12

A32f1n00 1000 24,510 32

A32f1n10 1000 24,120 32

A32f1n50 1000 22,794 32

Bpic2012A 13,087 60,849 10

Bpic2012 13,087 262,200 36

PAGE 6

Page 7: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Case Study – Settings – Replay results

Event log Running time (secs) CostsRepairExample 0.25 0.197

A32f1n00 11.00 0.000

A32f1n10 17.00 0.993

A32f1n50 32.00 4.521

Bpic2012A 0.59 1.293

Bpic2012 480.00 14.228

PAGE 7

Page 8: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Case Study – Decomp. Replay Results

Event log Passages Running time CostsRepairExample 6 0.43 171% 0.196 99%

A32f1n00 30 1.30 12% 0.000 100%

A32f1n10 30 1.10 7% 0.444 45%

A32f1n50 30 1.20 4% 2.155 48%

Bpic2012A 8 2.20 378% 0.629 49%

Bpic2012 12 DNF

PAGE 8

Page 9: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Bpic2012

PAGE 9

Page 10: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Bpic2012 – Problematic Passage

PAGE 10

Page 11: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

BPIC2012 – Attempt 1

PAGE 11

Running time Costs470.00 98% 8.722 61%

Page 12: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

BPIC2012 – Attempt 2

PAGE 12

Running time Costs190.00 40% 6.676 47%

Page 13: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

BPIC2012 – Attempt 3a

PAGE 13

Rule DescriptionR1 Incident arcs of places are equivalent

R2 Incident arcs of invisible transitions are equivalent

R3 Incident arcs of visible transitions with non-unique labels are equivalent to incident arcs of all transitions with the same label

R4 The i-th input arc of a visible transition with unique label is equivalent to the i-th output arc of that transition, if both exist

Page 14: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Case Study – Attempt 3a – Results

Event log Passages Running time CostsRepairExample 2 0.31 136% 0.197 100%

A32f1n00 4 1.90 18% 0.000 100%

A32f1n10 4 2.10 13% 0.929 94%

A32f1n50 4 3.10 10% 4.322 96%

Bpic2012A 1 0.74 125% 1,293 100%

Bpic2012 1 480.00 101% 14.228 100%

PAGE 14

Page 15: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

BPIC2012 – Attempt 3b

PAGE 15

Rule DescriptionR1 Incident arcs of places are equivalent

R2 The i-th input arc of an invisible transition is equivalent to the i-th output arc of that transition, it both exist

R3 Incident arcs of visible transitions with non-unique labels are equivalent to incident arcs of all transitions with the same label

R4 The i-th input arc of a visible transition with unique label is equivalent to the i-th output arc of that transition, if both exist

Page 16: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Case Study – Attempt 3b – Results

Event log Passages Running time CostsRepairExample 2 0.31 136% 0.197 100%

A32f1n00 4 1.90 18% 0.000 100%

A32f1n10 4 2.10 13% 0.929 94%

A32f1n50 4 3.10 10% 4.322 96%

Bpic2012A 3 2.30 391% 1.272 98%

Bpic2012 3 400.00 83% 14.227 100%

PAGE 16

Page 17: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

BPIC2012 – Attempt 3c

PAGE 17

Rule DescriptionR1 Incident arcs of places are equivalent

R2 The i-th input arc of an invisible non-milestone transition is equivalent to the i-th output arc of that transition, it both exist

R3 Incident arcs of visible transitions with non-unique labels are equivalent to incident arcs of all transitions with the same label

R4 The i-th input arc of a visible non-milestone transition with unique label is equivalent to the i-th output arc of that transition, if both exist

Page 18: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

BPIC2012 – Attempt 3c - Milestones

PAGE 18

Running time Costs100.00 22% 11.722 82%

Page 19: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Wrapping Up

Event log Passages Running time CostsA32f1n00 1

411.01.9 18%

0.0000.000 100%

A32f1n10 14

17.02.1 13%

0.9930.929 100%

A32f1n50 14

32.03.1 10%

4.5214.332 96%

Bpic2012 14

480.0100.0 22%

14.22811.722 82%

PAGE 19

Page 20: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Take-Home Points

• Decomposed Replay can be faster• At acceptable costs

• Decomposed Replay can be slower, much slower• What are good passages?

• Given the replayer we’re using• Chaining passages seems to help• Sparsest cuts through the net?

PAGE 20

Page 21: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Bpic2012 – Attempt 1

PAGE 21

Page 22: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Bpic2012 – Attempt 2

PAGE 22

Page 23: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Bpic2012 – Attempt 3c

PAGE 23

Subnet Running time Costs

A 100.0 7.102

B 66.0 0.951

C 2.8 0.863

D 97.0 2.806

Total 260.0 11.722

A

BC

D

Page 24: Decomposing Replay Problems: A Case Study Eric Verbeek and Wil van der Aalst

Tool Implementation – ProM 6

PAGE 24