Evaluating a Data Removal Strategy for Grid Environments Using Colored Petri Nets

1
Evaluating a Data Removal Strategy for Grid
Environments Using Colored Petri Nets

• Nikola Trcka, Wil van der Aalst, Carmen Bratosin,
  and Natalia Sidorova

2
Overview

• Modeling Grid Architectures using Colored Petri
  nets (CPNs)
• Motivation
• (Colored) Petri Nets
• Grid Architecture in Terms CPNs
• Data Removal Strategy
• Motivation
• Method and Evaluation
• Conclusions

3
Motivation

• There is little consensus on the definition of a
  grid
• Exact architecture hidden in code
• A good conceptual model for grids is needed!
• CPNs are formal, well established, very
  expressive, graphical, executable, and with
  excellent tool support (CPN Tools and PROM).

4
Petri nets

• Used to specify all kinds of systems (automotive,
  business, manufacturing, computer, etc.)
• Capable of modeling parallelism, choices,
  synchronization, message exchange, timing and
  resource sharing
• Business process example - complaints handling

5
Colored Petri nets

• CPNs add programming language aspects to Petri
  nets
• Tokens not just black but carry information
• Modeling technique
• Control-flow described by Petri nets
• Data and data manipulations described in ML
• Modularity via hierarchy
• Example

6
Grid Architecture (standard view)
Application layer
Middleware
Fabric layer
7
Grid architecture CPN model top page
8
It gets more complex inside - Scheduler
9
Why do we need a data removal strategy for grids?

• Grid data is typically large
• Grid storage space is usually limited
• Lack of space can cause delays
• Data often occupies grid storage space
  unnecessarily too long
• Data should be deleted as soon as possible!

10
Data removal method

• Aim Make the workflow delete data on time
• Method For every data element add a delete task
  at the exact place in the workflow from which
  this data is no longer needed
• Algorithm (per data element)
• Build the reachability graph of the workflow
  (state space)
• Find the last state from which the data element
  is not used anymore (can be more than one such
  state)
• Add a (prioritized) delete task, only enabled in
  this state

11
Applying the algorithm

• Simplified representation of a grid workflow
  enough
• Example
• a,b,c,d some data elements
• Data needs to be deleted
• a - immediately after the first task (p2)
• b - only at the end (p10)
• c - when in p5 and p6
• d - when in p8 or p9

12
Evaluation

• Application
• Small process mining experiment
• Testbed
• 9 nodes
• storage area
• 5000, 7000, and 10000
• processing cost
• 1, 5, and 10
• Min-Min scheduling
• varied period
• Varied arrival rate

13
Conclusions

• A reference grid architecture modeled in terms of
  CPNs
• CPNs can model all layers of the infrastructure
• The model is very descriptive and allows for
  efficient analysis
• A data removal strategy introduced
• Strategy evaluated by means of simulation, using
  the CPN model on a process mining experiment