Real-Time Software Transactional Memory: Contention Managers, Time Bounds, and Implementations

Transcription

1 Real-Time Software Transactional Memory: Contention Managers, Time Bounds, and Implementations Mohammed El-Shambakey Dissertation Submitted to the Faculty of the Virginia Polytechnic Institute and State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Computer Engineering Binoy Ravindran, Chair Robert P. Broadwater Cameron D. Patterson Mohamed Rizk Mohamed. Rizk Anil Kumar S. Vullikanti September 3, 23 Blacksburg, Virginia Keywords: Software Transactional Memory, Embedded Systems, Contention Managers Copyright 23, Mohammed El-Shambakey

2 Real-Time Software Transactional Memory: Contention Managers, Time Bounds, and Implementations Mohammed El-Shambakey (ABSTRACT)

3 Lock-based concurrency control suffers from programmability, scalability, and composability challenges. These challenges are exacerbated in emerging multicore architectures, on which improved software performance must be achieved by exposing greater concurrency. Transactional memory (TM) is an emerging alternative synchronization model for shared memory objects that promises to alleviate these difficulties. In this dissertation, we consider software transactional memory (STM) for concurrency control in multicore real-time software, and present a suite of real-time STM contention managers for resolving transactional conflicts. The contention managers are called, RCM,,, and. RCM and resolve conflicts using fixed and dynamic priorities of real-time tasks, respectively, and are naturally intended to be used with the fixed priority (e.g., G-RMA) and dynamic priority (e.g., G-EDF) multicore real-time schedulers, respectively. resolves conflicts based on task priorities as well as atomic section lengths, and can be used with G-EDF or G-RMA schedulers. Transactions under, RCM, and may retry due to conflicts with higher priority tasks even when there are no shared objects, i.e., transitive retry. avoids transitive retry and optimizes processor usage by lowering the priority of retrying transactions, thereby enabling other non-conflicting transactions to proceed., however, requires a priori knowledge of all requested objects for each atomic section, which is inconsistent with the semantics of dynamic STM. Moreover, its centralized design increases overhead. avoids transitive retry, do not require a priori knowledge of requested objects, and has a decentralized design. We establish upper bounds on transactional retry costs and task response times under the contention managers through schedulability analysis. Since and RCM preserve the semantics of the underlying real-time scheduler, their maximum transactional retry cost is double the maximum atomic section length. This is improved in the design of, which achieves shorter retry costs and tighter upper bounds. As avoids transitive retry and improves processor usage, it yields shorter retry costs and tighter upper bounds than, RCM, and. s upper bounds are similarly tight because it combines the advantages of and. We formally compare the proposed contention managers with each other, with lock-free synchronization, and with multiprocessor real-time locking protocols. Our analysis reveals that, for most cases,, RCM, and achieve higher schedulability than lock-free synchronization only when the atomic section length does not exceed half of lock-free synchronization s retry loop length. With equal periods and greater access times for shared objects, atomic section length under, RCM, and can be much larger than the retry loop length while still achieving better schedulability. With proper values for s design parameters, atomic section length can be larger than the retry loop length for better schedulability. Under, atomic section length can exceed lock-free s retry loop length and still achieve better schedulability in certain cases. achieves equal or better schedulability than lock-free with appropriate values for design parameters. The schedulability advantage of the contention managers over multiprocessor real-time locking protocols such as Global and depends upon the value of s max /L max, the ratio of the maxiii

4 imum transaction length to the maximum critical section length. s schedulability is equal or better than Global and if s max /L max 2. Checkpointing enables partial roll-back of transactions by recording transaction execution states (i.e., checkpoints) during execution, allowing roll-back to a previous checkpoint instead of transaction start, improving task response time. We extend with checkpointing and develop CP-, and identify the conditions under which CP- achieves equal or better schedulability than. We implement the contention managers in the Rochester STM framework and conduct experimental studies using a multicore real-time Linux kernel. Our studies reveal that among the contention managers, CP- has the best average-case performance. CP- s higher performance is due to the fact that s and s advantages are combined into the design of, which is the base of CP-. Moreover, checkpointing improves task response time. The contention managers were also found to have equal or better average-case performance than lock-free synchronization: more jobs meet their deadlines using CP-,, and than lock-free synchronization by 34.6%, 28.5%, and 32.4% (on average), respectively. The superiority of the contention managers is directly due to their better conflict resolution policies. Locking protocols such as and were found to perform better: more jobs meet their deadlines under and than any contention manager by 2.4% and 3.7% (on average), respectively. However, the proposed contention managers have numerous qualitative advantages over locking protocols. Locks do not compose, whereas STM transactions do. To allow multiple objects to be accessed in a critical section, assigns objects to non-conflicting groups, where each group is protected by a distinct lock. assumes that objects are accessed in a specific order to prevent deadlocks. In contrast, STM allows multiple objects to be accessed in a transaction in any order, while guaranteeing deadlock-freedom, which significantly increases programmability. Moreover, STM offers platform independence: the proposed contention managers can be entirely implemented in the user-space as a library. In contrast, real-time locking protocols such as and must be supported by the underlying platform (i.e., operating system or virtual machine). iv

5 Dedication To my parents, my wife, my daughter, and all my family v

6 Acknowledgments I would like to gratefully acknowledge the supervision of my advisor, Dr. Binoy Ravindran during my Ph.D. study, for his enthusiasm, and his great efforts to guide my research from the start. Many thanks to the rest of my committee: Dr. Robert P. Broadwater, Dr. Cameron D. Patterson, Dr. Mohamed R. Rizk, and Dr. Anil Vullikanti for their invaluable advice and comments during my preliminary and defence exams. It is a great honour to have them serving in my committee. In addition, I would like to thank all my previous and current colleagues in Systems Software Research Group, who provided my great environment for collaboration and discussion. Their warm suggestions and help made me never feel alone in this long journey. Last but not least, thank all my family members for their love and support. I am grateful to my parents, who always did their best in supporting my education from the childhood, and suggested me the correct direction to make my dream come true. Also, I thank my dear wife who devoted her love and support to me through the ups and downs over the past years. It is difficult to overstate my gratitude to her for being such a wonderful wife. This dissertation is dedicated to all the people who helped me and are helping me all the way. This work is supported in part by US National Science Foundation under grants CNS 95895, CNS 69, CNS 38, and CNS Any opinions, findings, and conclusions or recommendations expressed in this site are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. vi

7 Contents Introduction. Transactional Memory STM for Real-Time Software Research Contributions Organization Past and Related Work 9 2. Real-Time Locking Protocols Real-Time Lock-Free and Wait-Free Synchronization Real-Time Database Concurrency Control Real-Time TM Concurrency Control Models and Assumptions 25 4 The and RCM Contention Managers Illustrative Example Transitive Retry G-EDF Interference Retry Cost of Atomic Sections Upper Bound on Response Time RCM vii

8 4.2. Maximum Task Interference Retry Cost of Atomic Sections Upper Bound on Response Time Analytical Performance Comparison STM versus Lock-Free versus Lock-Free RCM versus Lock-Free STM versus Locking protocols Priority Inversion under versus RCM versus Priority Inversion under versus RCM versus Conclusions The Contention Manager Length-based CM Design and Rationale Illustrative Example Properties Retry Cost and Response Time of /G-EDF of /G-EDF /G-EDF versus /G-EDF versus Lock-free /G-EDF versus /G-EDF versus Retry Cost and Response Time of /G-RMA of /G-RMA viii

9 5.6. /G-RMA versus RCM /G-RMA versus Lock-free /G-RMA versus /G-RMA versus Conclusions The Contention Manager 7 6. Limitations of, RCM, and The Contention Manager Illustrative Example Properties Retry Cost and Response Time Under versus Competitors versus versus RCM versus /G-EDF versus /G-RMA versus Lock-free Synchronization versus Locking Protocols Conclusions The Contention Manager Motivation The Contention Manager Illustrative Example Retry Cost and Response Time Bounds Comparison versus versus RCM ix

10 7.4.3 versus /G-EDF versus G-RMA/ versus /G-EDF versus /G-RMA versus Lock-free versus Locking Protocols Conclusions Contention Manager with Checkpointing 8. Motivation Checkpointing (CP-) Checkpointing (CP) CP CP- Retry Cost CP- versus Conclusion Implementation and Experimental Evaluations 9. Methodology Tasksets Performance Measurements Results General results for General results for Avg RC Results Summary Qualitative Comparison Between STM, Locking Protocols and Lock-Free38. Compositionality Priority Inversion Nesting x

11 .4 Convoying Deadlock and Livelock Platform Dependence and Implementation Complexity Transparency Upper Bounds Complexity Memory Conclusion Conclusions and Future Work 47. Conclusions Future Work Bibliography 52 A Tasksets Properties 66 B Complete Deadline Satisfaction Results 94 C Complete Average Retry Cost Results 33 xi

12 List of Figures 4. Maximum interference of jobs of τ j to τ x i running on different processors, under G-EDF. T i = at j + b Effect of carried in job of τ j to retry cost of transactions in τ i Max interference of τ j to τ i in G-RMA Interference of s k i by various lengths of s l j for Tasksets 2, 542, 82 and for Tasksets 28, 568, 8 and for Tasksets 38, 578, 8 and for Tasksets 55, 595, 35 and for Tasksets 83, 623, 63 and for Tasksets 7, 647, 87 and for Tasksets, 65, 9 and for Tasksets 38, 678, 28 and for Tasksets 65, 75, 245 and for Tasksets 233, 773, 33 and for Tasksets 244, 784, 324 and for Tasksets 256, 796, 336 and for Tasksets 262, 82, 342 and Average RC for Tasksets 3, 543, 83 and Average RC for Tasksets 28, 568, 8 and Average RC for Tasksets 67, 67, 47 and xii

13 9.7 Average RC for Tasksets 8, 62, 6 and Average RC for Tasksets 97, 637, 77 and Average RC for Tasksets 6, 646, 86 and Average RC for Tasksets 3, 67, 2 and Average RC for Tasksets 39, 679, 29 and Average RC for Tasksets 58, 698, 238 and Average RC for Tasksets 78, 78, 258 and Average RC for Tasksets 85, 725, 265 and Average RC for Tasksets 2, 75, 29 and Average RC for Tasksets 234, 774, 34 and B. for Tasksets, 27, 54 and B.2 for Tasksets 2, 272, 542 and B.3 for Tasksets 3, 273, 543 and B.4 for Tasksets 4, 274, 544 and B.5 for Tasksets 5, 275, 545 and B.6 for Tasksets 6, 276, 546 and B.7 for Tasksets 7, 277, 547 and B.8 for Tasksets 8, 278, 548 and B.9 for Tasksets 9, 279, 549 and B. for Tasksets, 28, 55 and B. for Tasksets, 28, 55 and B.2 for Tasksets 2, 282, 552 and B.3 for Tasksets 3, 283, 553 and B.4 for Tasksets 4, 284, 554 and B.5 for Tasksets 5, 285, 555 and B.6 for Tasksets 6, 286, 556 and B.7 for Tasksets 7, 287, 557 and B.8 for Tasksets 8, 288, 558 and xiii

14 B.9 for Tasksets 9, 289, 559 and B.2 for Tasksets 2, 29, 56 and B.2 for Tasksets 2, 29, 56 and B.22 for Tasksets 22, 292, 562 and B.23 for Tasksets 23, 293, 563 and B.24 for Tasksets 24, 294, 564 and B.25 for Tasksets 25, 295, 565 and B.26 for Tasksets 26, 296, 566 and B.27 for Tasksets 27, 297, 567 and B.28 for Tasksets 28, 298, 568 and B.29 for Tasksets 29, 299, 569 and B.3 for Tasksets 3, 3, 57 and B.3 for Tasksets 3, 3, 57 and B.32 for Tasksets 32, 32, 572 and B.33 for Tasksets 33, 33, 573 and B.34 for Tasksets 34, 34, 574 and B.35 for Tasksets 35, 35, 575 and B.36 for Tasksets 36, 36, 576 and B.37 for Tasksets 37, 37, 577 and B.38 for Tasksets 38, 38, 578 and B.39 for Tasksets 39, 39, 579 and B.4 for Tasksets 4, 3, 58 and B.4 for Tasksets 4, 3, 58 and B.42 for Tasksets 42, 32, 582 and B.43 for Tasksets 43, 33, 583 and B.44 for Tasksets 44, 34, 584 and B.45 for Tasksets 45, 35, 585 and B.46 for Tasksets 46, 36, 586 and xiv

15 B.47 for Tasksets 47, 37, 587 and B.48 for Tasksets 48, 38, 588 and B.49 for Tasksets 49, 39, 589 and B.5 for Tasksets 5, 32, 59 and B.5 for Tasksets 5, 32, 59 and B.52 for Tasksets 52, 322, 592 and B.53 for Tasksets 53, 323, 593 and B.54 for Tasksets 54, 324, 594 and B.55 for Tasksets 55, 325, 595 and B.56 for Tasksets 56, 326, 596 and B.57 for Tasksets 57, 327, 597 and B.58 for Tasksets 58, 328, 598 and B.59 for Tasksets 59, 329, 599 and B.6 for Tasksets 6, 33, 6 and B.6 for Tasksets 6, 33, 6 and B.62 for Tasksets 62, 332, 62 and B.63 for Tasksets 63, 333, 63 and B.64 for Tasksets 64, 334, 64 and B.65 for Tasksets 65, 335, 65 and B.66 for Tasksets 66, 336, 66 and B.67 for Tasksets 67, 337, 67 and B.68 for Tasksets 68, 338, 68 and B.69 for Tasksets 69, 339, 69 and B.7 for Tasksets 7, 34, 6 and B.7 for Tasksets 7, 34, 6 and B.72 for Tasksets 72, 342, 62 and B.73 for Tasksets 73, 343, 63 and B.74 for Tasksets 74, 344, 64 and xv

16 B.75 for Tasksets 75, 345, 65 and B.76 for Tasksets 76, 346, 66 and B.77 for Tasksets 77, 347, 67 and B.78 for Tasksets 78, 348, 68 and B.79 for Tasksets 79, 349, 69 and B.8 for Tasksets 8, 35, 62 and B.8 for Tasksets 8, 35, 62 and B.82 for Tasksets 82, 352, 622 and B.83 for Tasksets 83, 353, 623 and B.84 for Tasksets 84, 354, 624 and B.85 for Tasksets 85, 355, 625 and B.86 for Tasksets 86, 356, 626 and B.87 for Tasksets 87, 357, 627 and B.88 for Tasksets 88, 358, 628 and B.89 for Tasksets 89, 359, 629 and B.9 for Tasksets 9, 36, 63 and B.9 for Tasksets 9, 36, 63 and B.92 for Tasksets 92, 362, 632 and B.93 for Tasksets 93, 363, 633 and B.94 for Tasksets 94, 364, 634 and B.95 for Tasksets 95, 365, 635 and B.96 for Tasksets 96, 366, 636 and B.97 for Tasksets 97, 367, 637 and B.98 for Tasksets 98, 368, 638 and B.99 for Tasksets 99, 369, 639 and B. for Tasksets, 37, 64 and B. for Tasksets, 37, 64 and B.2 for Tasksets 2, 372, 642 and xvi

17 B.3 for Tasksets 3, 373, 643 and B.4 for Tasksets 4, 374, 644 and B.5 for Tasksets 5, 375, 645 and B.6 for Tasksets 6, 376, 646 and B.7 for Tasksets 7, 377, 647 and B.8 for Tasksets 8, 378, 648 and B.9 for Tasksets 9, 379, 649 and B. for Tasksets, 38, 65 and B. for Tasksets, 38, 65 and B.2 for Tasksets 2, 382, 652 and B.3 for Tasksets 3, 383, 653 and B.4 for Tasksets 4, 384, 654 and B.5 for Tasksets 5, 385, 655 and B.6 for Tasksets 6, 386, 656 and B.7 for Tasksets 7, 387, 657 and B.8 for Tasksets 8, 388, 658 and B.9 for Tasksets 9, 389, 659 and B.2 for Tasksets 2, 39, 66 and B.2 for Tasksets 2, 39, 66 and B.22 for Tasksets 22, 392, 662 and B.23 for Tasksets 23, 393, 663 and B.24 for Tasksets 24, 394, 664 and B.25 for Tasksets 25, 395, 665 and B.26 for Tasksets 26, 396, 666 and B.27 for Tasksets 27, 397, 667 and B.28 for Tasksets 28, 398, 668 and B.29 for Tasksets 29, 399, 669 and B.3 for Tasksets 3, 4, 67 and xvii

18 B.3 for Tasksets 3, 4, 67 and B.32 for Tasksets 32, 42, 672 and B.33 for Tasksets 33, 43, 673 and B.34 for Tasksets 34, 44, 674 and B.35 for Tasksets 35, 45, 675 and B.36 for Tasksets 36, 46, 676 and B.37 for Tasksets 37, 47, 677 and B.38 for Tasksets 38, 48, 678 and B.39 for Tasksets 39, 49, 679 and B.4 for Tasksets 4, 4, 68 and B.4 for Tasksets 4, 4, 68 and B.42 for Tasksets 42, 42, 682 and B.43 for Tasksets 43, 43, 683 and B.44 for Tasksets 44, 44, 684 and B.45 for Tasksets 45, 45, 685 and B.46 for Tasksets 46, 46, 686 and B.47 for Tasksets 47, 47, 687 and B.48 for Tasksets 48, 48, 688 and B.49 for Tasksets 49, 49, 689 and B.5 for Tasksets 5, 42, 69 and B.5 for Tasksets 5, 42, 69 and B.52 for Tasksets 52, 422, 692 and B.53 for Tasksets 53, 423, 693 and B.54 for Tasksets 54, 424, 694 and B.55 for Tasksets 55, 425, 695 and B.56 for Tasksets 56, 426, 696 and B.57 for Tasksets 57, 427, 697 and B.58 for Tasksets 58, 428, 698 and xviii

19 B.59 for Tasksets 59, 429, 699 and B.6 for Tasksets 6, 43, 7 and B.6 for Tasksets 6, 43, 7 and B.62 for Tasksets 62, 432, 72 and B.63 for Tasksets 63, 433, 73 and B.64 for Tasksets 64, 434, 74 and B.65 for Tasksets 65, 435, 75 and B.66 for Tasksets 66, 436, 76 and B.67 for Tasksets 67, 437, 77 and B.68 for Tasksets 68, 438, 78 and B.69 for Tasksets 69, 439, 79 and B.7 for Tasksets 7, 44, 7 and B.7 for Tasksets 7, 44, 7 and B.72 for Tasksets 72, 442, 72 and B.73 for Tasksets 73, 443, 73 and B.74 for Tasksets 74, 444, 74 and B.75 for Tasksets 75, 445, 75 and B.76 for Tasksets 76, 446, 76 and B.77 for Tasksets 77, 447, 77 and B.78 for Tasksets 78, 448, 78 and B.79 for Tasksets 79, 449, 79 and B.8 for Tasksets 8, 45, 72 and B.8 for Tasksets 8, 45, 72 and B.82 for Tasksets 82, 452, 722 and B.83 for Tasksets 83, 453, 723 and B.84 for Tasksets 84, 454, 724 and B.85 for Tasksets 85, 455, 725 and B.86 for Tasksets 86, 456, 726 and xix

20 B.87 for Tasksets 87, 457, 727 and B.88 for Tasksets 88, 458, 728 and B.89 for Tasksets 89, 459, 729 and B.9 for Tasksets 9, 46, 73 and B.9 for Tasksets 9, 46, 73 and B.92 for Tasksets 92, 462, 732 and B.93 for Tasksets 93, 463, 733 and B.94 for Tasksets 94, 464, 734 and B.95 for Tasksets 95, 465, 735 and B.96 for Tasksets 96, 466, 736 and B.97 for Tasksets 97, 467, 737 and B.98 for Tasksets 98, 468, 738 and B.99 for Tasksets 99, 469, 739 and B.2 for Tasksets 2, 47, 74 and B.2 for Tasksets 2, 47, 74 and B.22 for Tasksets 22, 472, 742 and B.23 for Tasksets 23, 473, 743 and B.24 for Tasksets 24, 474, 744 and B.25 for Tasksets 25, 475, 745 and B.26 for Tasksets 26, 476, 746 and B.27 for Tasksets 27, 477, 747 and B.28 for Tasksets 28, 478, 748 and B.29 for Tasksets 29, 479, 749 and B.2 for Tasksets 2, 48, 75 and B.2 for Tasksets 2, 48, 75 and B.22 for Tasksets 22, 482, 752 and B.23 for Tasksets 23, 483, 753 and B.24 for Tasksets 24, 484, 754 and xx

21 B.25 for Tasksets 25, 485, 755 and B.26 for Tasksets 26, 486, 756 and B.27 for Tasksets 27, 487, 757 and B.28 for Tasksets 28, 488, 758 and B.29 for Tasksets 29, 489, 759 and B.22 for Tasksets 22, 49, 76 and B.22 for Tasksets 22, 49, 76 and B.222 for Tasksets 222, 492, 762 and B.223 for Tasksets 223, 493, 763 and B.224 for Tasksets 224, 494, 764 and B.225 for Tasksets 225, 495, 765 and B.226 for Tasksets 226, 496, 766 and B.227 for Tasksets 227, 497, 767 and B.228 for Tasksets 228, 498, 768 and B.229 for Tasksets 229, 499, 769 and B.23 for Tasksets 23, 5, 77 and B.23 for Tasksets 23, 5, 77 and B.232 for Tasksets 232, 52, 772 and B.233 for Tasksets 233, 53, 773 and B.234 for Tasksets 234, 54, 774 and B.235 for Tasksets 235, 55, 775 and B.236 for Tasksets 236, 56, 776 and B.237 for Tasksets 237, 57, 777 and B.238 for Tasksets 238, 58, 778 and B.239 for Tasksets 239, 59, 779 and B.24 for Tasksets 24, 5, 78 and B.24 for Tasksets 24, 5, 78 and B.242 for Tasksets 242, 52, 782 and xxi

22 B.243 for Tasksets 243, 53, 783 and B.244 for Tasksets 244, 54, 784 and B.245 for Tasksets 245, 55, 785 and B.246 for Tasksets 246, 56, 786 and B.247 for Tasksets 247, 57, 787 and B.248 for Tasksets 248, 58, 788 and B.249 for Tasksets 249, 59, 789 and B.25 for Tasksets 25, 52, 79 and B.25 for Tasksets 25, 52, 79 and B.252 for Tasksets 252, 522, 792 and B.253 for Tasksets 253, 523, 793 and B.254 for Tasksets 254, 524, 794 and B.255 for Tasksets 255, 525, 795 and B.256 for Tasksets 256, 526, 796 and B.257 for Tasksets 257, 527, 797 and B.258 for Tasksets 258, 528, 798 and B.259 for Tasksets 259, 529, 799 and B.26 for Tasksets 26, 53, 8 and B.26 for Tasksets 26, 53, 8 and B.262 for Tasksets 262, 532, 82 and B.263 for Tasksets 263, 533, 83 and B.264 for Tasksets 264, 534, 84 and B.265 for Tasksets 265, 535, 85 and B.266 for Tasksets 266, 536, 86 and B.267 for Tasksets 267, 537, 87 and B.268 for Tasksets 268, 538, 88 and B.269 for Tasksets 269, 539, 89 and B.27 for Tasksets 27, 54, 8 and xxii

23 C. Avg RC for Tasksets, 27, 54 and C.2 Avg RC for Tasksets 2, 272, 542 and C.3 Avg RC for Tasksets 3, 273, 543 and C.4 Avg RC for Tasksets 4, 274, 544 and C.5 Avg RC for Tasksets 5, 275, 545 and C.6 Avg RC for Tasksets 6, 276, 546 and C.7 Avg RC for Tasksets 7, 277, 547 and C.8 Avg RC for Tasksets 8, 278, 548 and C.9 Avg RC for Tasksets 9, 279, 549 and C. Avg RC for Tasksets, 28, 55 and C. Avg RC for Tasksets, 28, 55 and C.2 Avg RC for Tasksets 2, 282, 552 and C.3 Avg RC for Tasksets 3, 283, 553 and C.4 Avg RC for Tasksets 4, 284, 554 and C.5 Avg RC for Tasksets 5, 285, 555 and C.6 Avg RC for Tasksets 6, 286, 556 and C.7 Avg RC for Tasksets 7, 287, 557 and C.8 Avg RC for Tasksets 8, 288, 558 and C.9 Avg RC for Tasksets 9, 289, 559 and C.2 Avg RC for Tasksets 2, 29, 56 and C.2 Avg RC for Tasksets 2, 29, 56 and C.22 Avg RC for Tasksets 22, 292, 562 and C.23 Avg RC for Tasksets 23, 293, 563 and C.24 Avg RC for Tasksets 24, 294, 564 and C.25 Avg RC for Tasksets 25, 295, 565 and C.26 Avg RC for Tasksets 26, 296, 566 and C.27 Avg RC for Tasksets 27, 297, 567 and C.28 Avg RC for Tasksets 28, 298, 568 and xxiii

24 C.29 Avg RC for Tasksets 29, 299, 569 and C.3 Avg RC for Tasksets 3, 3, 57 and C.3 Avg RC for Tasksets 3, 3, 57 and C.32 Avg RC for Tasksets 32, 32, 572 and C.33 Avg RC for Tasksets 33, 33, 573 and C.34 Avg RC for Tasksets 34, 34, 574 and C.35 Avg RC for Tasksets 35, 35, 575 and C.36 Avg RC for Tasksets 36, 36, 576 and C.37 Avg RC for Tasksets 37, 37, 577 and C.38 Avg RC for Tasksets 38, 38, 578 and C.39 Avg RC for Tasksets 39, 39, 579 and C.4 Avg RC for Tasksets 4, 3, 58 and C.4 Avg RC for Tasksets 4, 3, 58 and C.42 Avg RC for Tasksets 42, 32, 582 and C.43 Avg RC for Tasksets 43, 33, 583 and C.44 Avg RC for Tasksets 44, 34, 584 and C.45 Avg RC for Tasksets 45, 35, 585 and C.46 Avg RC for Tasksets 46, 36, 586 and C.47 Avg RC for Tasksets 47, 37, 587 and C.48 Avg RC for Tasksets 48, 38, 588 and C.49 Avg RC for Tasksets 49, 39, 589 and C.5 Avg RC for Tasksets 5, 32, 59 and C.5 Avg RC for Tasksets 5, 32, 59 and C.52 Avg RC for Tasksets 52, 322, 592 and C.53 Avg RC for Tasksets 53, 323, 593 and C.54 Avg RC for Tasksets 54, 324, 594 and C.55 Avg RC for Tasksets 55, 325, 595 and C.56 Avg RC for Tasksets 56, 326, 596 and xxiv

25 C.57 Avg RC for Tasksets 57, 327, 597 and C.58 Avg RC for Tasksets 58, 328, 598 and C.59 Avg RC for Tasksets 59, 329, 599 and C.6 Avg RC for Tasksets 6, 33, 6 and C.6 Avg RC for Tasksets 6, 33, 6 and C.62 Avg RC for Tasksets 62, 332, 62 and C.63 Avg RC for Tasksets 63, 333, 63 and C.64 Avg RC for Tasksets 64, 334, 64 and C.65 Avg RC for Tasksets 65, 335, 65 and C.66 Avg RC for Tasksets 66, 336, 66 and C.67 Avg RC for Tasksets 67, 337, 67 and C.68 Avg RC for Tasksets 68, 338, 68 and C.69 Avg RC for Tasksets 69, 339, 69 and C.7 Avg RC for Tasksets 7, 34, 6 and C.7 Avg RC for Tasksets 7, 34, 6 and C.72 Avg RC for Tasksets 72, 342, 62 and C.73 Avg RC for Tasksets 73, 343, 63 and C.74 Avg RC for Tasksets 74, 344, 64 and C.75 Avg RC for Tasksets 75, 345, 65 and C.76 Avg RC for Tasksets 76, 346, 66 and C.77 Avg RC for Tasksets 77, 347, 67 and C.78 Avg RC for Tasksets 78, 348, 68 and C.79 Avg RC for Tasksets 79, 349, 69 and C.8 Avg RC for Tasksets 8, 35, 62 and C.8 Avg RC for Tasksets 8, 35, 62 and C.82 Avg RC for Tasksets 82, 352, 622 and C.83 Avg RC for Tasksets 83, 353, 623 and C.84 Avg RC for Tasksets 84, 354, 624 and xxv

26 C.85 Avg RC for Tasksets 85, 355, 625 and C.86 Avg RC for Tasksets 86, 356, 626 and C.87 Avg RC for Tasksets 87, 357, 627 and C.88 Avg RC for Tasksets 88, 358, 628 and C.89 Avg RC for Tasksets 89, 359, 629 and C.9 Avg RC for Tasksets 9, 36, 63 and C.9 Avg RC for Tasksets 9, 36, 63 and C.92 Avg RC for Tasksets 92, 362, 632 and C.93 Avg RC for Tasksets 93, 363, 633 and C.94 Avg RC for Tasksets 94, 364, 634 and C.95 Avg RC for Tasksets 95, 365, 635 and C.96 Avg RC for Tasksets 96, 366, 636 and C.97 Avg RC for Tasksets 97, 367, 637 and C.98 Avg RC for Tasksets 98, 368, 638 and C.99 Avg RC for Tasksets 99, 369, 639 and C.Avg RC for Tasksets, 37, 64 and C.Avg RC for Tasksets, 37, 64 and C.2Avg RC for Tasksets 2, 372, 642 and C.3Avg RC for Tasksets 3, 373, 643 and C.4Avg RC for Tasksets 4, 374, 644 and C.5Avg RC for Tasksets 5, 375, 645 and C.6Avg RC for Tasksets 6, 376, 646 and C.7Avg RC for Tasksets 7, 377, 647 and C.8Avg RC for Tasksets 8, 378, 648 and C.9Avg RC for Tasksets 9, 379, 649 and C.Avg RC for Tasksets, 38, 65 and C.Avg RC for Tasksets, 38, 65 and C.2Avg RC for Tasksets 2, 382, 652 and xxvi

27 C.3Avg RC for Tasksets 3, 383, 653 and C.4Avg RC for Tasksets 4, 384, 654 and C.5Avg RC for Tasksets 5, 385, 655 and C.6Avg RC for Tasksets 6, 386, 656 and C.7Avg RC for Tasksets 7, 387, 657 and C.8Avg RC for Tasksets 8, 388, 658 and C.9Avg RC for Tasksets 9, 389, 659 and C.2Avg RC for Tasksets 2, 39, 66 and C.2Avg RC for Tasksets 2, 39, 66 and C.22Avg RC for Tasksets 22, 392, 662 and C.23Avg RC for Tasksets 23, 393, 663 and C.24Avg RC for Tasksets 24, 394, 664 and C.25Avg RC for Tasksets 25, 395, 665 and C.26Avg RC for Tasksets 26, 396, 666 and C.27Avg RC for Tasksets 27, 397, 667 and C.28Avg RC for Tasksets 28, 398, 668 and C.29Avg RC for Tasksets 29, 399, 669 and C.3Avg RC for Tasksets 3, 4, 67 and C.3Avg RC for Tasksets 3, 4, 67 and C.32Avg RC for Tasksets 32, 42, 672 and C.33Avg RC for Tasksets 33, 43, 673 and C.34Avg RC for Tasksets 34, 44, 674 and C.35Avg RC for Tasksets 35, 45, 675 and C.36Avg RC for Tasksets 36, 46, 676 and C.37Avg RC for Tasksets 37, 47, 677 and C.38Avg RC for Tasksets 38, 48, 678 and C.39Avg RC for Tasksets 39, 49, 679 and C.4Avg RC for Tasksets 4, 4, 68 and xxvii

28 C.4Avg RC for Tasksets 4, 4, 68 and C.42Avg RC for Tasksets 42, 42, 682 and C.43Avg RC for Tasksets 43, 43, 683 and C.44Avg RC for Tasksets 44, 44, 684 and C.45Avg RC for Tasksets 45, 45, 685 and C.46Avg RC for Tasksets 46, 46, 686 and C.47Avg RC for Tasksets 47, 47, 687 and C.48Avg RC for Tasksets 48, 48, 688 and C.49Avg RC for Tasksets 49, 49, 689 and C.5Avg RC for Tasksets 5, 42, 69 and C.5Avg RC for Tasksets 5, 42, 69 and C.52Avg RC for Tasksets 52, 422, 692 and C.53Avg RC for Tasksets 53, 423, 693 and C.54Avg RC for Tasksets 54, 424, 694 and C.55Avg RC for Tasksets 55, 425, 695 and C.56Avg RC for Tasksets 56, 426, 696 and C.57Avg RC for Tasksets 57, 427, 697 and C.58Avg RC for Tasksets 58, 428, 698 and C.59Avg RC for Tasksets 59, 429, 699 and C.6Avg RC for Tasksets 6, 43, 7 and C.6Avg RC for Tasksets 6, 43, 7 and C.62Avg RC for Tasksets 62, 432, 72 and C.63Avg RC for Tasksets 63, 433, 73 and C.64Avg RC for Tasksets 64, 434, 74 and C.65Avg RC for Tasksets 65, 435, 75 and C.66Avg RC for Tasksets 66, 436, 76 and C.67Avg RC for Tasksets 67, 437, 77 and C.68Avg RC for Tasksets 68, 438, 78 and xxviii

29 C.69Avg RC for Tasksets 69, 439, 79 and C.7Avg RC for Tasksets 7, 44, 7 and C.7Avg RC for Tasksets 7, 44, 7 and C.72Avg RC for Tasksets 72, 442, 72 and C.73Avg RC for Tasksets 73, 443, 73 and C.74Avg RC for Tasksets 74, 444, 74 and C.75Avg RC for Tasksets 75, 445, 75 and C.76Avg RC for Tasksets 76, 446, 76 and C.77Avg RC for Tasksets 77, 447, 77 and C.78Avg RC for Tasksets 78, 448, 78 and C.79Avg RC for Tasksets 79, 449, 79 and C.8Avg RC for Tasksets 8, 45, 72 and C.8Avg RC for Tasksets 8, 45, 72 and C.82Avg RC for Tasksets 82, 452, 722 and C.83Avg RC for Tasksets 83, 453, 723 and C.84Avg RC for Tasksets 84, 454, 724 and C.85Avg RC for Tasksets 85, 455, 725 and C.86Avg RC for Tasksets 86, 456, 726 and C.87Avg RC for Tasksets 87, 457, 727 and C.88Avg RC for Tasksets 88, 458, 728 and C.89Avg RC for Tasksets 89, 459, 729 and C.9Avg RC for Tasksets 9, 46, 73 and C.9Avg RC for Tasksets 9, 46, 73 and C.92Avg RC for Tasksets 92, 462, 732 and C.93Avg RC for Tasksets 93, 463, 733 and C.94Avg RC for Tasksets 94, 464, 734 and C.95Avg RC for Tasksets 95, 465, 735 and C.96Avg RC for Tasksets 96, 466, 736 and xxix

30 C.97Avg RC for Tasksets 97, 467, 737 and C.98Avg RC for Tasksets 98, 468, 738 and C.99Avg RC for Tasksets 99, 469, 739 and C.2Avg RC for Tasksets 2, 47, 74 and C.2Avg RC for Tasksets 2, 47, 74 and C.22Avg RC for Tasksets 22, 472, 742 and C.23Avg RC for Tasksets 23, 473, 743 and C.24Avg RC for Tasksets 24, 474, 744 and C.25Avg RC for Tasksets 25, 475, 745 and C.26Avg RC for Tasksets 26, 476, 746 and C.27Avg RC for Tasksets 27, 477, 747 and C.28Avg RC for Tasksets 28, 478, 748 and C.29Avg RC for Tasksets 29, 479, 749 and C.2Avg RC for Tasksets 2, 48, 75 and C.2Avg RC for Tasksets 2, 48, 75 and C.22Avg RC for Tasksets 22, 482, 752 and C.23Avg RC for Tasksets 23, 483, 753 and C.24Avg RC for Tasksets 24, 484, 754 and C.25Avg RC for Tasksets 25, 485, 755 and C.26Avg RC for Tasksets 26, 486, 756 and C.27Avg RC for Tasksets 27, 487, 757 and C.28Avg RC for Tasksets 28, 488, 758 and C.29Avg RC for Tasksets 29, 489, 759 and C.22Avg RC for Tasksets 22, 49, 76 and C.22Avg RC for Tasksets 22, 49, 76 and C.222Avg RC for Tasksets 222, 492, 762 and C.223Avg RC for Tasksets 223, 493, 763 and C.224Avg RC for Tasksets 224, 494, 764 and xxx

31 C.225Avg RC for Tasksets 225, 495, 765 and C.226Avg RC for Tasksets 226, 496, 766 and C.227Avg RC for Tasksets 227, 497, 767 and C.228Avg RC for Tasksets 228, 498, 768 and C.229Avg RC for Tasksets 229, 499, 769 and C.23Avg RC for Tasksets 23, 5, 77 and C.23Avg RC for Tasksets 23, 5, 77 and C.232Avg RC for Tasksets 232, 52, 772 and C.233Avg RC for Tasksets 233, 53, 773 and C.234Avg RC for Tasksets 234, 54, 774 and C.235Avg RC for Tasksets 235, 55, 775 and C.236Avg RC for Tasksets 236, 56, 776 and C.237Avg RC for Tasksets 237, 57, 777 and C.238Avg RC for Tasksets 238, 58, 778 and C.239Avg RC for Tasksets 239, 59, 779 and C.24Avg RC for Tasksets 24, 5, 78 and C.24Avg RC for Tasksets 24, 5, 78 and C.242Avg RC for Tasksets 242, 52, 782 and C.243Avg RC for Tasksets 243, 53, 783 and C.244Avg RC for Tasksets 244, 54, 784 and C.245Avg RC for Tasksets 245, 55, 785 and C.246Avg RC for Tasksets 246, 56, 786 and C.247Avg RC for Tasksets 247, 57, 787 and C.248Avg RC for Tasksets 248, 58, 788 and C.249Avg RC for Tasksets 249, 59, 789 and C.25Avg RC for Tasksets 25, 52, 79 and C.25Avg RC for Tasksets 25, 52, 79 and C.252Avg RC for Tasksets 252, 522, 792 and xxxi

32 C.253Avg RC for Tasksets 253, 523, 793 and C.254Avg RC for Tasksets 254, 524, 794 and C.255Avg RC for Tasksets 255, 525, 795 and C.256Avg RC for Tasksets 256, 526, 796 and C.257Avg RC for Tasksets 257, 527, 797 and C.258Avg RC for Tasksets 258, 528, 798 and C.259Avg RC for Tasksets 259, 529, 799 and C.26Avg RC for Tasksets 26, 53, 8 and C.26Avg RC for Tasksets 26, 53, 8 and C.262Avg RC for Tasksets 262, 532, 82 and C.263Avg RC for Tasksets 263, 533, 83 and C.264Avg RC for Tasksets 264, 534, 84 and C.265Avg RC for Tasksets 265, 535, 85 and C.266Avg RC for Tasksets 266, 536, 86 and C.267Avg RC for Tasksets 267, 537, 87 and C.268Avg RC for Tasksets 268, 538, 88 and C.269Avg RC for Tasksets 269, 539, 89 and C.27Avg RC for Tasksets 27, 54, 8 and xxxii