Carnegie Mellon University. Hard Real-Time Multiprocessor Scheduling (Part II) Marcus Völp

Size: px

Start display at page:

Download "Carnegie Mellon University. Hard Real-Time Multiprocessor Scheduling (Part II) Marcus Völp"

Alvin Lucas
6 years ago
Views:

1 Hard eal-time Multiprocessor Scheduling (Part II) Marcus Völp

2 Lessons Learned from UP two concepts to counteract priority inversion priority inheritance priority ceiling priority ceiling priority inheritance L L L U L U 12/11/2013 Dr.-Ing. Marcus Völp 45

3 Locking Protocols Where to execute the critical section? local vs. remote [Brandenburg 10, Brandenburg 13] How to order conflicting critical sections? FIFO vs. priorities How unrelated tasks affect the lock holders? Whether and how unrelated tasks are affected? 12/11/2013 Dr.-Ing. Marcus Völp 46

4 Where to execute? local vs. remote [Brandenburg 10, Brandenburg 13] 12/11/2013 Dr.-Ing. Marcus Völp 47

5 Where to execute? local vs. remote [Brandenburg 10, Brandenburg 13] 12/11/2013 Dr.-Ing. Marcus Völp 48

6 Where to execute? local vs. remote [Brandenburg 10, Brandenburg 13] 12/11/2013 Dr.-Ing. Marcus Völp 50

7 How to order conflicts? FIFO vs. priorities [Brandenburg 10, Brandenburg 13] L L FIFO: O(n) Priorities: O(higher prioritized on other CPUs) O(1)? O(m)? 12/11/2013 Dr.-Ing. Marcus Völp 51

8 Unrelated tasks affect the lock Block local vs. remote [Brandenburg 10, Brandenburg 13] ( ) C 12/11/2013 Dr.-Ing. Marcus Völp 52

9 Unrelated tasks affect the lock Block local vs. remote [Brandenburg 10, Brandenburg 13] ( ) C 12/11/2013 Dr.-Ing. Marcus Völp 53

10 How unrelated threads are affected Block local vs. remote [Brandenburg 10, Brandenburg 13] ( ) C 12/11/2013 Dr.-Ing. Marcus Völp 54

11 How unrelated threads are affected Block local vs. remote [Brandenburg 10, Brandenburg 13] ( ) C 12/11/2013 Dr.-Ing. Marcus Völp 55

12 T-Locking Terminology schedulability analysis suspension-aware suspension-oblivious blocking priority-inversion blocking deferral blocking 12/11/2013 Dr.-Ing. Marcus Völp 57

13 T-Locking Terminology schedulability analysis suspension-aware suspension-oblivious blocking priority-inversion blocking deferral blocking 12/11/2013 Dr.-Ing. Marcus Völp 58

14 euse UP Protocols? Distributed Priority-Ceiling Protocol (DPCP) [ajkuma, Sha, Lehocky 88] 12/11/2013 Dr.-Ing. Marcus Völp 59

15 euse UP Protocols? Distributed Priority-Ceiling Protocol (DPCP) [ajkuma, Sha, Lehocky 88] 12/11/2013 Dr.-Ing. Marcus Völp 60

16 euse UP Protocols? Distributed Priority-Ceiling Protocol (DPCP) [ajkuma, Sha, Lehocky 88] synchronization processor DPCP = BPCP m+1 12/11/2013 Dr.-Ing. Marcus Völp 61

17 euse UP Protocols? Multiprocessor Stack esource Protocol (MSP) [Gai, Lipari, Natale 01] synchronization processor MSP = MSP m+1 12/11/2013 Dr.-Ing. Marcus Völp 62

Björn Brandenburg Improved Analysis and Evaluation of eal-time Semaphore Protocols for Partitioned Fixed Priority Scheduling 13 MPCP [ajkumar 90, Lakshmanan 09] FMLP [Brandenburg 11] DPCP [ajkumar

18 Björn Brandenburg Improved Analysis and Evaluation of eal-time Semaphore Protocols for Partitioned Fixed Priority Scheduling 13 MPCP [ajkumar 90, Lakshmanan 09] FMLP [Brandenburg 11] DPCP [ajkumar 88] DFLP [Brandenburg 11] hdwallpaperscool.com wait queue priority FIFO priority FIFO protocol type shared mem. (local) shared mem. (local) distributed (sync. CPU) distributed (sync. CPU) asymptotically optimal (wrt. maximal blocking) No Yes No Yes 12/11/2013 Dr.-Ing. Marcus Völp 63

19 Björn Brandenburg Improved Analysis and Evaluation of eal-time Semaphore Protocols for Partitioned Fixed Priority Scheduling 13 hdwallpaperscool.com 12/11/2013 Dr.-Ing. Marcus Völp 64

20 Caches QPI QPI hdwallpaperscool.com L3 L3 AM AM 12/11/2013 Dr.-Ing. Marcus Völp 65

21 Caches 201 cycles [y: OSDI 08] 3 cycles QPI L3 Snoop filter 50 cycles reduce coherent traffic by keeping track of cached lines inclusive L3: line, => line L3 snoop filter: AM keep address not data 12/11/2013 Dr.-Ing. Marcus Völp 66

22 Caches [y: OSDI 08] reduce coherent traffic by keeping track of cached lines inclusive L3: line, => line L3 snoop filter: keep address not data 12/11/2013 Dr.-Ing. Marcus Völp 67

23 T - Locks for Clusters eal-time esource-sharing under Clustered Scheduling: Mutex, eader-writer, and k-exclusion Locks Björn Brandenburg, Jim Anderson 11 hdwallpaperscool.com What if every task helps out a little bit when it is released? only the n-highest prioritized tasks may acquire locks? 12/11/2013 Dr.-Ing. Marcus Völp 68

24 T - Locks for Clusters J 1 eal-time esource-sharing under Clustered Scheduling: Mutex, eader-writer, and k-exclusion Locks Björn Brandenburg, Jim Anderson 11 L hdwallpaperscool.com J 2 J 3 L J 4 J 5 J 6 L 12/11/2013 Dr.-Ing. Marcus Völp 69

25 T - Locks for Clusters eal-time esource-sharing under Clustered Scheduling: Mutex, eader-writer, and k-exclusion Locks Björn Brandenburg, Jim Anderson 11 hdwallpaperscool.com D1. J i may issue a request only if it among the n = CPU highest prioritized jobs D2. When released at time ta, Jd becomes Ji s priority donor if a) J i was the c = Cluster highestpriority pending job released prior to J d, b) J d has one of the c highest base priorities, and c) J i has issued a request that is incomplete D3. J i inherits J d s priority D4. If J d ceases to be among the c highest-priority jobs in its cluster, the newly released job takes over the donor role D5. / D6. schedule either J i or its donor J d D7. a priority donor may not issue requests (J d suspends in this case while being a priority donor) D8. If J d finishes execution while being a donor, J d remains the donor but is suspended D9. J d stops being a donor if a) J i completes its request, b) J i becomes one of the c highest prioritized jobs c) another job takes over being the donor of J i 12/11/2013 Dr.-Ing. Marcus Völp 70

26 Outline Last Week: Terminology and Notations Anomalies and Impossibility esults Partitioned Scheduling Global Scheduling Optimal MP Scheduling Practical Matters Today: Exercise: UP esource Protocols esources Caches / Memories One More esource Dynamic Tasks Peek and Poke into other Bleeding Edge esearch 12/11/2013 Dr.-Ing. Marcus Völp 71

27 Caches A Coordinated Approach for Practical OS-Level Cache Management in Multi- eal-time Systems Hyoseung Kim, Arvind Kandhalu, agunathan (aj) ajkumar (ECTS 13) hdwallpaperscool.com L3 L3 40% slowdown 27% slowdown PASEC on i7 12/11/2013 Dr.-Ing. Marcus Völp 72

28 Caches A Coordinated Approach for Practical OS-Level Cache Management in Multi- eal-time Systems Hyoseung Kim, Arvind Kandhalu, agunathan (aj) ajkumar (ECTS 13) hdwallpaperscool.com L3 AM 12/11/2013 Dr.-Ing. Marcus Völp 73

29 Dynamic Tasks Algorithm and complexity for the global scheduling of sporadic tasks on multiprocessors with work-limited parallelism S. Collette, L. Cucu, J. Goossens [eal-time and Network Systems, 07] hdwallpaperscool.com J (C, P, Γ ) Γ (γ,, γ,,, γ, ) Job J i with j CPUs completes in i,j C i 12/11/2013 Dr.-Ing. Marcus Völp 74

30 Dynamic Tasks / untime Threads Threadpools Apple s Grand Central Dispatch, C++ Futures, Scala Actors, hdwallpaperscool.com 12/11/2013 Dr.-Ing. Marcus Völp 75

31 Outline Last Week: Terminology and Notations Anomalies and Impossibility esults Partitioned Scheduling Global Scheduling Optimal MP Scheduling Practical Matters Today: Exercise: UP esource Protocols esources Caches / Memories One More esource Dynamic Tasks Peek and Poke into other Bleeding Edge esearch 12/11/2013 Dr.-Ing. Marcus Völp 76

32 eferences [Burns '09]. Davis, A. Burns, A Survey of Hard eal-time Scheduling Algorithms and Schedulability Analysis Techniques for Multiprocessor Systems, 2009 [Colette] S. Collette, L. Cucu, J. Goossens, Algorithm and complexity for the global scheduling of sporadic tasks on multiprocessors with worklimited parallelism, Int. Conf. On eal-time and Network Systems, 2007 [Levin '10] G. Levin, S. Funk, C.Sadowski, I. Pye, S. Brandt, DP-Fair: A Simple Model for Understanding Optimal Multiprocessor Scheduling, 2010 [Hong '88] K. Hong, J. Leung, On-line scheduling of real-time tasks, TSS, 1988 [Anderson '01] B. Anderson, S. Baruah, J. Jonsson, Static-priority scheduling on multiprocessors, TSS, 2001 [Lakshmanan '09] [Edmonds] K. Lakshmanan,. ajkumar, J. Lehoczky, Partitioned Fixed-Priority J. Edmonds, K. Pruhs, Scalably scheduling processes with arbitrary Preemptive Scheduling for Multi- Processors, ECTS, 2009 speedup curves, Symp. On Discrete Algorithms, /11/2013 Dr.-Ing. Marcus Völp 77

33 eferences [Fischer '07] N. Fischer, The multiprocessor real-time scheduling of general task systems, PhD. Thesis, of North Carolina at Chapel Hill, 2007 [Dertouzos '89] M. Dertouzos, A. Mok, Multiprocessor scheduling in a hard real-time environment, IEEE Trans. on Software Engineering, 1989 [Dhall] S. Dhall, C, Liu, On a eal-time Scheduling Problem, Operations esearch, 1978 [Baruah '06] S. Baruah, A. Burns, Sustainable Scheduling Analysis, TSS, 2006 [y '08] S. Boyd-Wickizer, H. Chen,. Chen, Y. Mao, F. Kaashoek,. Morris, A. Pesterev, L. Stein, M. Wu, Y. Dai, Y. Zhang, Z. Zhang, y: An Operating System for Many s, OSDI 2008 [Carpenter '04] J. Carpenter, S. Funk, P. Holman, A. Srinivasan, J. Anderson, S. Baruah, A categorization of real-time multiprocessor scheduling problems and algorithms, Handbook of Scheduling: Algorithms, Models, and Performance Analysis, 2004 [Whitehead] J. Leung, J. Whitehead, On the Complexity of Fixed-Priority Scheduling of Periodic, eal-time Tasks,1982 [Brandenburg '08] B. Brandenburg, J. Calandrino, A. Block, H. Leontyev, J. Anderson, eal-time Synchronization on Multiprocessors: To Block or Not to Block, to Suspend or Spin, TAS, /11/2013 Dr.-Ing. Marcus Völp 78

34 eferences [Gai '03] P. Gai, M. Natale, G. Lipari, A. Ferrari, C. Gabellini, P. Marceca, A comparison of MPCP and MSP when sharing resources in the Janus multiple-processor on a chip platform, TAS, 2003 [Lauzac '98] S. Lauzac,. Melhem, D. Mosse, Comparison of global and partitioning schemes for scheduling rate monotonic tasks on a multiprocessor, EuroMicro, 1998 [Block] [Baruah '96] A. Block, H. Leontyev, B. Brandenburg, J. Anderson, A Flexible eal- S. Baruah, N. Cohen, G. Plaxton, D. Varvel, Proportionate progress: A Time Locking Protocol for Multiprocessors, TCSA, 2007 notion of fairness in resource allocation, Algorithmica 15, 1996 [MSP] P. Gai, G. Lipari, M. Natale, Minimizing memory utilization of realtime task sets in single and multi-processor systems-on-a-chip, TSS, 2001 [Brandenburg'11] B.Brandenburg, J. Anderson, eal-time esource-sharing under Clustered Scheduling: Mutex, eader-writer, and k-exclusion Locks, ESOICS 2011 [MPCP]. ajkumar, L. Sha, J. Lehoczky, eal-time synchronization protocols for multiprocessors, TSS, /11/2013 Dr.-Ing. Marcus Völp 79

35 eferences [Brandenburg '13] B. Brandenburg, Improved Analysis and Evaluation of eal-time Semaphore Protocols for Partitioned Fixed Priority Scheduling, TAS, 2013 [Lauzac '98] S. Lauzac,. Melhem, D. Mosse, Comparison of global and [Brandenburg 10] B. Brandenburg, J. Anderson, Optimality esults for Multiprocessor eal-time Locking, TSS, /11/2013 Dr.-Ing. Marcus Völp 80

Real-Time Systems. Hard Real-Time Multiprocessor Scheduling

Real-Time Systems. Hard Real-Time Multiprocessor Scheduling Real-Time Systems Hard Real-Time Multiprocessor Scheduling Marcus Völp WS 2015/16 Outline Introduction Terminology, Notation and Assumptions Anomalies + Impossibility Results Partitioned Scheduling (no