1. Grid-enabled Probabilistic Model Checking with PRISM Yi Zhang, David Parker, Marta Kwiatkowska University of Birmingham

2. Outline Introduction –PRISM –Parallel Numerical Engines Integrate Parallel Numerical Engines into PRISM Conclusion

3. PRISM A probabilistic model checker Supports three types of models: –discrete-time Markov chains (DTMCs) –continuous-time Markov chains (CTMCs) –Markov decision processes (MDPs) A wide range of properties of these can be analysed.

4. PRISM Case Studies Randomised distributed algorithms Communication and multimedia protocols –Bluetooth, FireWire, Zeroconf Biological process modelling Security Protocols Dynamic Power Management www.cs.bham.ac.uk/~dxp/prism/casestudies

5. Motivation Probabilistic Model Checking is a powerful method. But because of state space explosion problem, it can be very expensive. Two approaches to combat the problem: –Symbolic approaches, such as MTBDD structures. –Parallel and distributed solutions for probabilistic model checking. How to integrate parallel numerical engines into PRISM.

6. Parallel Numerical Engines A Parallel Gauss-Seidel Iterative Method for shared memory machines. A Parallel Wavefront Guass-Seidel Method for Message passing machines. Based on MTBDD data structures Solving linear equation systems for analysis of CTMC and DTMC

7. Integrate Parallel Numerical Engines into PRISM Manage remote computation resources for end users. Free end users from learning remote scheduling systems. Handling data transfer on behalf of end users. Monitoring job execution on remote computation resources.

8. The Role of Globus Toolkits Provide building blocks for our middleware. –GSI for security –GRAM for job management –GSI-Openssh for file transfering –Grid services for data handling and job monitoring.

9. Structures of Grid-enabled PRISM

10. Job Submission Component Based on WS-GRAM. Generates job description files Communicates with WS-GRAM services at remote resources.

11. Data Transfer Using GSI-OpenSSH for file transfer –Matrices –Vectors Create grid services for fine-grained data access –Block by block

12. Job Monitoring Based on WS-GRAM at current stage. –Provide basic information about job status Grid service for job monitoring is under development. –Information about job status –Runtime information –Convergence rate information

13. Examples

15. Conclusion A grid-based middleware for PRISM. Provide easy access of remote parallel computation resources for end users. A foundation for future parallelisation work in PRISM.