Multiple processor systems
4 likes2,469 views
The document is about an advanced operating systems course for fall 2012 taught by Van Hoang Nguyen. It discusses topics like communication techniques for multiprocessors, multicomputers and distributed systems. It also covers OS code, scheduling, deadlocks, message passing performance, and processor allocation algorithms for multiprocessors, multicomputers and distributed systems.
Multiple processor systems
- 1. Dept. of Computer Science – FITA – HUA Van Hoang Nguyen Mail: startnewday85@gmail.com Department of Computer Science – FITA – HUA Advanced Operating System Course ---------------------------------- Fall 2012
- 2. Advanced Operating System – Fall 2012 Van Hoang Nguyen
- 3. Advanced Operating System – Fall 2012 › Communication Techniques › Multiprocessor › Multicomputer › Distributed System Van Hoang Nguyen
- 4. Advanced Operating System – Fall 2012 OS 1 OS 2 OS 3 OS 4 Van Hoang Nguyen
- 5. Advanced Operating System – Fall 2012 OS Code Van Hoang Nguyen
- 6. Advanced Operating System – Fall 2012 OS Code › static memory allocation › no data sharing › no parallel applications Van Hoang Nguyen
- 7. Advanced Operating System – Fall 2012 OS Van Hoang Nguyen
- 8. Advanced Operating System – Fall 2012 OS Van Hoang Nguyen
- 9. Advanced Operating System – Fall 2012 OS › balances processes and memory dynamically › eliminates the master CPU bottleneck Van Hoang Nguyen
- 10. Advanced Operating System – Fall 2012 Dead Lock problems Van Hoang Nguyen
- 11. Advanced Operating System – Fall 2012 Not work • TSL instruction is non-trivial in SMPs Van Hoang Nguyen
- 12. Advanced Operating System – Fall 2012 • Busy waiting/spin lock › Cache thrashing Spinning to wait for a lock is not always a bad idea Van Hoang Nguyen
- 13. Advanced Operating System – Fall 2012 • Spinning vs. Switching › Scheduling critical section may be held › if critical sections are short spinning may be better than blocking › static analysis of critical section duration can determine whether to spin or block › dynamic analysis can improve performance Van Hoang Nguyen
- 14. Advanced Operating System – Fall 2012 • Single processor scheduling › Which thread should be run next? ---- Time Sharing •Multiprocessor scheduling › Which thread should be run next? --- Time Sharing › Which CPU to run it on? --- Space Sharing Van Hoang Nguyen
- 15. Advanced Operating System – Fall 2012 • Time sharing approach • Space sharing approach Van Hoang Nguyen
- 16. Advanced Operating System – Fall 2012 • Time sharing approach Van Hoang Nguyen
- 17. Advanced Operating System – Fall 2012 • Message passing performance Van Hoang Nguyen
- 18. Advanced Operating System – Fall 2012 • Message passing performance Van Hoang Nguyen
- 19. Advanced Operating System – Fall 2012 • Message passing performance Sacrifice protection for performance Van Hoang Nguyen
- 20. Advanced Operating System – Fall 2012 • Programming model Van Hoang Nguyen
- 21. Advanced Operating System – Fall 2012 • Programming model Van Hoang Nguyen
- 22. Advanced Operating System – Fall 2012 • Programming model › Send with copy › User-level interrupt › Copy on write Van Hoang Nguyen
- 23. Advanced Operating System – Fall 2012 • Programming model › False sharing › Consistency Van Hoang Nguyen
- 24. Advanced Operating System – Fall 2012 • Multiprocessor • Multicomputer Van Hoang Nguyen
- 25. Advanced Operating System – Fall 2012 • Processor allocation Algorithms’ characteristics Van Hoang Nguyen
- 26. Advanced Operating System – Fall 2012 • A graph-theoretic deterministic algorithm Van Hoang Nguyen
- 27. Advanced Operating System – Fall 2012 • Sender-initiated distributed heuristic algorithm Van Hoang Nguyen
- 28. Advanced Operating System – Fall 2012 • Receiver-initiated distributed heuristic algorithm Van Hoang Nguyen
- 29. Advanced Operating System – Fall 2012 Comparing multiprocessors, multicomputers, distributed systems Item Multiprocessor Multicomputer Distributed system Node configuration CPU CPU, RAM, IC Complete computer Node peripherals All shared Shared exc. May be Full set per node disk Location Same rack Same room Possibly world wide Inter-node Shared RAM Dedicated Traditional network communication interconnect Operating systems One, shared Multiple, same Possibly all different File systems One, shared One, shared Each node has own Administration One organization One organization Many organizations Van Hoang Nguyen
- 30. Advanced Operating System – Fall 2012 Van Hoang Nguyen
- 31. Advanced Operating System – Fall 2012 Van Hoang Nguyen