a scalable architecture for ordered parallelism
play

A Scalable Architecture for Ordered Parallelism Mark Jeffrey , - PowerPoint PPT Presentation

Jul 22, 2023 •120 likes •1.22k views

A Scalable Architecture for Ordered Parallelism Mark Jeffrey , Suvinay Subramanian, Cong Yan, Joel Emer, Daniel Sanchez MICRO 2015 Multicores Target Easy Parallelism 2 Multicores Target Easy Parallelism 2 Regular : known tasks and data

  1. Insights about Ordered Parallelism 6 1. With perfect speculation, parallelism is plentiful Parallelism Ideal schedule E max 800x A C B D window=64 26x B D E window=1k 180x 2. Tasks are tiny : 32 instructions on average 3. Independent tasks are far away in program order Can execute N tasks ahead A C B D E of the earliest active task N -task window Need a large window of speculation

  2. Prior Work Can’t Mine Ordered Parallelism 7

  3. Prior Work Can’t Mine Ordered Parallelism 7 ¨ Thread-Level Speculation (TLS) parallelizes loops and function calls in sequential programs

  4. Prior Work Can’t Mine Ordered Parallelism 7 ¨ Thread-Level Speculation (TLS) parallelizes loops and function calls in sequential programs Max parallelism TLS parallelism 800x 1.1x

  5. Prior Work Can’t Mine Ordered Parallelism 7 ¨ Thread-Level Speculation (TLS) parallelizes loops and function calls in sequential programs Execution order ≠ creation order Max parallelism TLS parallelism 800x 1.1x

  6. Prior Work Can’t Mine Ordered Parallelism 7 ¨ Thread-Level Speculation (TLS) parallelizes loops and function calls in sequential programs Execution order ≠ creation order Max parallelism TLS parallelism Task-scheduling priority queues 800x 1.1x introduce false data dependences

  7. Prior Work Can’t Mine Ordered Parallelism 7 ¨ Thread-Level Speculation (TLS) parallelizes loops and function calls in sequential programs Execution order ≠ creation order Max parallelism TLS parallelism Task-scheduling priority queues 800x 1.1x introduce false data dependences ¨ Sophisticated parallel algorithms yield limited speedup

  8. Prior Work Can’t Mine Ordered Parallelism 7 ¨ Thread-Level Speculation (TLS) parallelizes loops and function calls in sequential programs Execution order ≠ creation order Max parallelism TLS parallelism Task-scheduling priority queues 800x 1.1x introduce false data dependences ¨ Sophisticated parallel algorithms yield limited speedup bfs sssp astar msf des silo 64 Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c

  9. Swarm Mines Ordered Parallelism 8 bfs sssp astar msf des silo 64 117x Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c

  10. Swarm Mines Ordered Parallelism 8 bfs sssp astar msf des silo 64 117x Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c

  11. Swarm Mines Ordered Parallelism 8 bfs sssp astar msf des silo 64 117x Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c ¨ Execution model based on timestamped tasks

  12. Swarm Mines Ordered Parallelism 8 bfs sssp astar msf des silo 64 117x Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c ¨ Execution model based on timestamped tasks ¨ Architecture executes tasks speculatively out of order ¤ Leverages execution model to scale

  13. Outline 9 ¨ Understanding Ordered Parallelism ¨ Swarm ¨ Evaluation

  14. Swarm Execution Model 10 Programs consist of timestamped tasks

  15. Swarm Execution Model 10 Programs consist of timestamped tasks ¤ Tasks can create children tasks with >= timestamp ¤ Tasks appear to execute in timestamp order

  16. Swarm Execution Model 10 Programs consist of timestamped tasks ¤ Tasks can create children tasks with >= timestamp ¤ Tasks appear to execute in timestamp order ¤ Programmed with implicitly-parallel task API swarm::enqueue(fptr, ¡ts, ¡args...); ¡ 0 2 4 5 6 3 4 7

  17. Swarm Execution Model 10 Programs consist of timestamped tasks ¤ Tasks can create children tasks with >= timestamp ¤ Tasks appear to execute in timestamp order ¤ Programmed with implicitly-parallel task API swarm::enqueue(fptr, ¡ts, ¡args...); ¡ 0 2 4 5 6 3 4 7 Conveys new work to hardware as soon as possible

  18. Swarm Execution Model 10 Programs consist of timestamped tasks ¤ Tasks can create children tasks with >= timestamp ¤ Tasks appear to execute in timestamp order ¤ Programmed with implicitly-parallel task API swarm::enqueue(fptr, ¡ts, ¡args...); ¡ 0 2 4 5 6 3 4 7 Conveys new work to hardware as soon as possible

  19. Swarm Task Example: Dijkstra 11 void ¡ ssspTask(Timestamp ¡dist, ¡Vertex& ¡v) ¡{ ¡ ¡ ¡if ¡(!v.isVisited()) ¡{ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡v.distance ¡= ¡dist; ¡ ¡ ¡ ¡ ¡ for ¡(Vertex& ¡u ¡: ¡v.neighbors) ¡{ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡Timestamp ¡uDist ¡= ¡dist ¡+ ¡edgeWeight(v, ¡u); ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡swarm::enqueue(&ssspTask, ¡uDist, ¡u); ¡ ¡ ¡ ¡ ¡} ¡ ¡ ¡} ¡ } ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡

  20. Swarm Task Example: Dijkstra 11 void ¡ ssspTask(Timestamp ¡dist, ¡Vertex& ¡v) ¡{ ¡ ¡ ¡if ¡(!v.isVisited()) ¡{ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡v.distance ¡= ¡dist; ¡ ¡ ¡ ¡ ¡ for ¡(Vertex& ¡u ¡: ¡v.neighbors) ¡{ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡Timestamp ¡uDist ¡= ¡dist ¡+ ¡edgeWeight(v, ¡u); ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡swarm::enqueue(&ssspTask, ¡uDist, ¡u); ¡ ¡ ¡ ¡ ¡} ¡ ¡ ¡} ¡ } ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡

  21. Swarm Task Example: Dijkstra 11 void ¡ ssspTask(Timestamp ¡dist, ¡Vertex& ¡v) ¡{ ¡ ¡ ¡if ¡(!v.isVisited()) ¡{ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡v.distance ¡= ¡dist; ¡ ¡ ¡ ¡ ¡ for ¡(Vertex& ¡u ¡: ¡v.neighbors) ¡{ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡Timestamp ¡uDist ¡= ¡dist ¡+ ¡edgeWeight(v, ¡u); ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡swarm::enqueue(&ssspTask, ¡uDist, ¡u); ¡ ¡ ¡ ¡ ¡} ¡ ¡ ¡} ¡ Timestamp } ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡

  22. Swarm Task Example: Dijkstra 11 void ¡ ssspTask(Timestamp ¡dist, ¡Vertex& ¡v) ¡{ ¡ ¡ ¡if ¡(!v.isVisited()) ¡{ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡v.distance ¡= ¡dist; ¡ ¡ ¡ ¡ ¡ for ¡(Vertex& ¡u ¡: ¡v.neighbors) ¡{ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡Timestamp ¡uDist ¡= ¡dist ¡+ ¡edgeWeight(v, ¡u); ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡swarm::enqueue(&ssspTask, ¡uDist, ¡u); ¡ ¡ ¡ ¡ ¡} ¡ ¡ ¡} ¡ Timestamp } ¡ ¡ swarm::enqueue(ssspTask, ¡0, ¡sourceVertex); ¡ swarm::run(); ¡

  23. Swarm Architecture Overview 12 Tiled Multicore Tile Organization Memory controller L3 Cache Bank Router Memory controller Memory controller L2 Cache Tile L1I/D L1I/D L1I/D L1I/D Core Core Core Core Task Unit Memory controller

  24. Swarm Architecture Overview 12 Tiled Multicore Tile Organization Task Unit Memory controller L3 Cache Bank Router Memory controller Memory controller L2 Cache TQ Tile L1I/D L1I/D L1I/D L1I/D Core Core Core Core Task Unit CQ Memory controller Per-tile task units: ¨ Task Queue : holds task descriptors ¨ Commit Queue : holds speculative state of finished tasks

  25. Swarm Architecture Overview 12 Tiled Multicore Tile Organization Task Unit Memory controller L3 Cache Bank Router Memory controller Memory controller L2 Cache TQ Tile L1I/D L1I/D L1I/D L1I/D Core Core Core Core Task Unit CQ Memory controller Per-tile task units: ¨ Task Queue : holds task descriptors ¨ Commit Queue : holds speculative state of finished tasks Commit queues provide the window of speculation

  26. Task Unit Queues 13 ¨ Task queue : holds task descriptors ¨ Commit Queue : holds speculative state of finished tasks 68 Task States: IDLE (I) RUNNING (R) FINISHED (F) Task Queue Cores Commit Queue 9, I 2 10, I 2, R 8 8, R 3 3, F

  27. Task Unit Queues 13 ¨ Task queue : holds task descriptors ¨ Commit Queue : holds speculative state of finished tasks 69 Task States: IDLE (I) RUNNING (R) FINISHED (F) Task Queue Cores Commit Queue New Task ( timestamp=7, 7, I taskFn, args ) 9, I 2 10, I 2, R 8 8, R 3 3, F

  28. Task Unit Queues 14 ¨ Task queue : holds task descriptors ¨ Commit Queue : holds speculative state of finished tasks 70 Task States: IDLE (I) RUNNING (R) FINISHED (F) Task Queue Cores Commit Queue 7, R 9, I 7 10, I 2, F 2 8 8, R 3 3, F

  29. Task Unit Queues 15 ¨ Task queue : holds task descriptors ¨ Commit Queue : holds speculative state of finished tasks 71 Task States: IDLE (I) RUNNING (R) FINISHED (F) Task Queue Cores Commit Queue 7, F 9, I 8 10, I 7 2, F 2 8 8, R 3 3, F

  30. Task Unit Queues 16 ¨ Task queue : holds task descriptors ¨ Commit Queue : holds speculative state of finished tasks 72 Task States: IDLE (I) RUNNING (R) FINISHED (F) Task Queue Cores Commit Queue 7, F 9, R 9 9 10, I 7 2, F 2 8 8, R 3 3, F

  31. Task Unit Queues 16 ¨ Task queue : holds task descriptors ¨ Commit Queue : holds speculative state of finished tasks 73 Task States: IDLE (I) RUNNING (R) FINISHED (F) Task Queue Cores Commit Queue 7, F 9, R 9 9 10, I 7 2, F 2 8 8, R 3 3, F Similar to a reorder buffer, but at the task level

  32. High-Throughput Ordered Commits 17 ¨ Suppose 64-cycle tasks execute on 64 cores ¤ 1 task commit/cycle to scale ¤ TLS commit schemes (successor lists, commit token) too slow

  33. High-Throughput Ordered Commits 17 ¨ Suppose 64-cycle tasks execute on 64 cores ¤ 1 task commit/cycle to scale ¤ TLS commit schemes (successor lists, commit token) too slow ¨ We adapt “Virtual Time” [Jefferson, TOPLAS 1985] Tile Tile Tile … 1 2 N GVT Arbiter

  34. High-Throughput Ordered Commits 17 ¨ Suppose 64-cycle tasks execute on 64 cores ¤ 1 task commit/cycle to scale ¤ TLS commit schemes (successor lists, commit token) too slow ¨ We adapt “Virtual Time” [Jefferson, TOPLAS 1985] ¨ Tiles periodically communicate to Tile Tile Tile … 1 2 N find the earliest unfinished task GVT Arbiter

  35. High-Throughput Ordered Commits 17 ¨ Suppose 64-cycle tasks execute on 64 cores ¤ 1 task commit/cycle to scale ¤ TLS commit schemes (successor lists, commit token) too slow ¨ We adapt “Virtual Time” [Jefferson, TOPLAS 1985] ¨ Tiles periodically communicate to Tile Tile Tile … 1 2 N find the earliest unfinished task GVT Arbiter

  36. High-Throughput Ordered Commits 17 ¨ Suppose 64-cycle tasks execute on 64 cores ¤ 1 task commit/cycle to scale ¤ TLS commit schemes (successor lists, commit token) too slow ¨ We adapt “Virtual Time” [Jefferson, TOPLAS 1985] ¨ Tiles periodically communicate to Tile Tile Tile … 1 2 N find the earliest unfinished task ¨ Tiles commit all tasks that precede it GVT Arbiter

  37. High-Throughput Ordered Commits 17 ¨ Suppose 64-cycle tasks execute on 64 cores ¤ 1 task commit/cycle to scale ¤ TLS commit schemes (successor lists, commit token) too slow ¨ We adapt “Virtual Time” [Jefferson, TOPLAS 1985] ¨ Tiles periodically communicate to Tile Tile Tile … 1 2 N find the earliest unfinished task ¨ Tiles commit all tasks that precede it GVT Arbiter With large commit queues, many tasks commit at once

  38. High-Throughput Ordered Commits 17 ¨ Suppose 64-cycle tasks execute on 64 cores ¤ 1 task commit/cycle to scale ¤ TLS commit schemes (successor lists, commit token) too slow ¨ We adapt “Virtual Time” [Jefferson, TOPLAS 1985] ¨ Tiles periodically communicate to Tile Tile Tile … 1 2 N find the earliest unfinished task ¨ Tiles commit all tasks that precede it GVT Arbiter With large commit queues, many tasks commit at once Amortizes commit costs among many tasks

  39. Speculative Execution Example 18 0 Timestamp order Core 0 0 Core 1 Core 2 Time

  40. Speculative Execution Example 18 0 1 3 Timestamp order Core 0 0 1 Core 1 3 Core 2 Time

  41. Speculative Execution Example 18 0 1 3 Timestamp order Core 0 0 1 Core 1 3 Core 2 Time ¨ Tasks can execute even if parent is still speculative ¤ Uncovers more parallelism

  42. Speculative Execution Example 18 0 1 4 3 5 Timestamp order Core 0 0 1 Core 1 3 4 Core 2 5 Time ¨ Tasks can execute even if parent is still speculative ¤ Uncovers more parallelism

  43. Speculative Execution Example 18 0 1 2 4 3 5 Timestamp order Core 0 0 1 2 Core 1 3 4 Core 2 5 Time ¨ Tasks can execute even if parent is still speculative ¤ Uncovers more parallelism

  44. Speculative Execution Example 18 0 1 2 Data dependence 4 3 5 Timestamp order Core 0 0 1 2 Core 1 3 4 Core 2 5 Time ¨ Tasks can execute even if parent is still speculative ¤ Uncovers more parallelism

  45. Speculative Execution Example 18 0 1 2 Data dependence 4 3 5 Timestamp order Core 0 0 1 2 Core 1 3 4 Core 2 5 Time ¨ Tasks can execute even if parent is still speculative ¤ Uncovers more parallelism ¤ May trigger cascading (but selective) aborts

  46. Speculative Execution Example 18 0 1 2 Data dependence 4 3 5 Timestamp order Core 0 0 1 2 Core 1 3 4 Core 2 5 Time ¨ Tasks can execute even if parent is still speculative ¤ Uncovers more parallelism ¤ May trigger cascading (but selective) aborts

  47. Swarm Speculation Mechanisms 19 ¨ Key requirements for speculative execution: ¤ Fast commits ¤ Large speculative window à Small per-task speculative state

  48. Swarm Speculation Mechanisms 19 ¨ Key requirements for speculative execution: ¤ Fast commits ¤ Large speculative window à Small per-task speculative state ¨ Eager versioning + timestamp-based conflict detection ¤ Bloom filters for cheap read/write sets [Yen, HPCA 2007]

  49. Swarm Speculation Mechanisms 19 ¨ Key requirements for speculative execution: ¤ Fast commits ¤ Large speculative window à Small per-task speculative state ¨ Eager versioning + timestamp-based conflict detection ¤ Bloom filters for cheap read/write sets [Yen, HPCA 2007] ¤ Uses hierarchical memory system to filter conflict checks

  50. Swarm Speculation Mechanisms 19 ¨ Key requirements for speculative execution: ¤ Fast commits ¤ Large speculative window à Small per-task speculative state ¨ Eager versioning + timestamp-based conflict detection ¤ Bloom filters for cheap read/write sets [Yen, HPCA 2007] ¤ Uses hierarchical memory system to filter conflict checks ¨ Enables two helpful properties Forwarding of still-speculative data 1. On rollback, corrective writes abort dependent tasks only 2.

  51. Outline 20 ¨ Understanding Ordered Parallelism ¨ Swarm ¨ Evaluation

  52. Evaluation Methodology 21 ¨ Event-driven, sequential, Pin-based simulator ¨ Target system: 64-core, 16-tile chip Memory controller Router L3 Cache Bank 16 MB shared L3 (1MB/tile) Memory controller Memory controller 256 KB per-tile L2s L2 Cache Tile 32 KB per-core L1s L1I/D L1I/D L1I/D L1I/D Core Core Core Core 4096 task queue entries (64/core) Task Unit 1024 commit queue entries (16/core) Memory controller 256-byte, 8-way Bloom filters

  53. Evaluation Methodology 21 ¨ Event-driven, sequential, Pin-based simulator ¨ Target system: 64-core, 16-tile chip Memory controller Router L3 Cache Bank 16 MB shared L3 (1MB/tile) Memory controller Memory controller 256 KB per-tile L2s L2 Cache Tile 32 KB per-core L1s L1I/D L1I/D L1I/D L1I/D Core Core Core Core 4096 task queue entries (64/core) Task Unit 1024 commit queue entries (16/core) Memory controller 256-byte, 8-way Bloom filters

  54. Evaluation Methodology 21 ¨ Event-driven, sequential, Pin-based simulator ¨ Target system: 64-core, 16-tile chip Memory controller Router L3 Cache Bank 16 MB shared L3 (1MB/tile) Memory controller Memory controller 256 KB per-tile L2s L2 Cache Tile 32 KB per-core L1s L1I/D L1I/D L1I/D L1I/D Core Core Core Core 4096 task queue entries (64/core) Task Unit 1024 commit queue entries (16/core) Memory controller 256-byte, 8-way Bloom filters ¨ Scalability experiments from 1-64 cores ¤ Scaled-down systems have fewer tiles

  55. Swarm vs. Software Versions 22 bfs sssp astar msf des silo 64 117x Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c

  56. Swarm vs. Software Versions 22 bfs sssp astar msf des silo 64 117x Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 43x – 117x faster than serial versions

  57. Swarm vs. Software Versions 22 bfs sssp astar msf des silo 64 117x Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 43x – 117x faster than serial versions 3x – 18x faster than parallel versions

  58. Swarm vs. Software Versions 22 bfs sssp astar msf des silo 64 117x Speedup 32 1 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 1c 32c 64c 43x – 117x faster than serial versions 3x – 18x faster than parallel versions Simple implicitly-parallel code

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Harmonizing Speculative and Non-Speculative Execution in Architectures for Ordered Parallelism

Harmonizing Speculative and Non-Speculative Execution in Architectures for Ordered Parallelism

Harmonizing Speculative and Non-Speculative Execution in Architectures for Ordered Parallelism MA MARK C. JEFFR FFREY , VICTOR A. YING, SUVINAY SUBRAMANIAN, HYUN RYONG LEE, JOEL EMER, DANIEL SANCHEZ MI MICRO 2018 There is a (false)

1.48k views • 130 slides
Parallel architectures Electronic Computers LM Parallelism 1 Architecture Architecture:

Parallel architectures Electronic Computers LM Parallelism 1 Architecture Architecture:

Parallel architectures Electronic Computers LM Parallelism 1 Architecture Architecture: functional behaviour of a computer. For instance a processor which executes DLX code Implementation: a logical network implementing the

1.38k views • 103 slides
Hierarchical Phasers for Scalable Synchronization and Reductions in Dynamic Parallelism IPDPS

Hierarchical Phasers for Scalable Synchronization and Reductions in Dynamic Parallelism IPDPS

Hierarchical Phasers for Scalable Synchronization and Reductions in Dynamic Parallelism IPDPS 2010 April 22nd, 2010 Jun Shirako and Vivek Sarkar Rice University Introduction Major crossroads in computer industry Processor clock speeds are

723 views • 43 slides
Unit 8: Superscalar Pipelines Then: Static & dynamic scheduling Extract much more

Unit 8: Superscalar Pipelines Then: Static & dynamic scheduling Extract much more

A Key Theme: Parallelism Previously: pipeline-level parallelism Work on execute of one instruction in parallel with decode of next CIS 501: Computer Architecture Next: instruction-level parallelism (ILP) Execute multiple independent

441 views • 7 slides
Data-Parallel Architectures Nima Honarmand Spring 2016 :: CSE 502 Computer Architecture

Data-Parallel Architectures Nima Honarmand Spring 2016 :: CSE 502 Computer Architecture

Spring 2016 :: CSE 502 Computer Architecture Data-Parallel Architectures Nima Honarmand Spring 2016 :: CSE 502 Computer Architecture Overview Data Parallelism vs. Control (Thread-Level) Parallelism Data Parallelism: parallelism

434 views • 31 slides
Automation of Determination of Optimal Intra-Compute Node Parallelism PRESENTED BY: Scalable

Automation of Determination of Optimal Intra-Compute Node Parallelism PRESENTED BY: Scalable

Automation of Determination of Optimal Intra-Compute Node Parallelism PRESENTED BY: Scalable Tools Workshop Antonio Gmez agomez@tacc.utexas.edu James C. Browne 8/1/16 1 Why? Many applications using MPI for intra-node

487 views • 17 slides
WaveScalar Dataflow machine good at exploiting ILP dataflow parallelism + traditional

WaveScalar Dataflow machine good at exploiting ILP dataflow parallelism + traditional

WaveScalar Dataflow machine good at exploiting ILP dataflow parallelism + traditional coarser-grain parallelism cheap thread management memory ordering enforced through wave-ordered memory Winter 2006 CSE 548 - WaveScalar 1

579 views • 27 slides
Exploitation of instruction level parallelism Computer Architecture J. Daniel Garca Snchez

Exploitation of instruction level parallelism Computer Architecture J. Daniel Garca Snchez

Exploitation of instruction level parallelism Exploitation of instruction level parallelism Computer Architecture J. Daniel Garca Snchez (coordinator) David Expsito Singh Francisco Javier Garca Blas ARCOS Group Computer Science and

689 views • 55 slides
BEYOND FLUX BEYOND FLUX SCALABLE FRONTEND ARCHITECTURES SCALABLE FRONTEND ARCHITECTURES USING

BEYOND FLUX BEYOND FLUX SCALABLE FRONTEND ARCHITECTURES SCALABLE FRONTEND ARCHITECTURES USING

BEYOND FLUX BEYOND FLUX SCALABLE FRONTEND ARCHITECTURES SCALABLE FRONTEND ARCHITECTURES USING PUBLISH/SUBSCRIBE USING PUBLISH/SUBSCRIBE Michael Kurze @ goto Amsterdam 2016 1 WHY ARCHITECTURE? WHY ARCHITECTURE? IT'S ABOUT TIME! IT'S ABOUT

520 views • 38 slides
A simple, scalable app architecture with Android annotations Luke Sleeman Freelance Android

A simple, scalable app architecture with Android annotations Luke Sleeman Freelance Android

A simple, scalable app architecture with Android annotations Luke Sleeman Freelance Android developer lukesleeman.com.au Image CC: https://flic.kr/p/6oqCZB Agenda Introduction The architecture - an overview Services Domain

588 views • 39 slides
Inside the Machine: An Introduction to Architecture, Parallelism & its Applications

Inside the Machine: An Introduction to Architecture, Parallelism & its Applications

Inside the Machine: An Introduction to Architecture, Parallelism & its Applications Soumyabrata Dev The ADAPT SFI Research Centre https://soumyabrata.github.io/ UCD - Beijing-Dublin International College (BDIC) 7-May-2019 The ADAPT Centre

667 views • 18 slides
Superscalar Organization Nima Honarmand Spring 2016 :: CSE 502 Computer Architecture

Superscalar Organization Nima Honarmand Spring 2016 :: CSE 502 Computer Architecture

Spring 2016 :: CSE 502 Computer Architecture Superscalar Organization Nima Honarmand Spring 2016 :: CSE 502 Computer Architecture Instruction-Level Parallelism (ILP) Recall: Parallelism is the number of independent tasks

481 views • 11 slides
CS654 Advanced Computer Architecture Lec 8 Instruction Level Parallelism Peter Kemper

CS654 Advanced Computer Architecture Lec 8 Instruction Level Parallelism Peter Kemper

CS654 Advanced Computer Architecture Lec 8 Instruction Level Parallelism Peter Kemper Adapted from the slides of EECS 252 by Prof. David Patterson Electrical Engineering and Computer Sciences University of California, Berkeley Review from

656 views • 40 slides
WSO2 Message Broker Scalable persistent Messaging System Outline Messaging Scalable

WSO2 Message Broker Scalable persistent Messaging System Outline Messaging Scalable

WSO2 Message Broker Scalable persistent Messaging System Outline Messaging Scalable Messaging Distributed Message Brokers WSO2 MB Architecture o Distributed Pub/sub architecture o Distributed Queues architecture User Story

1.12k views • 33 slides
Reagents: Functional programming meets scalable concurrency Aaron Turon Northeastern University

Reagents: Functional programming meets scalable concurrency Aaron Turon Northeastern University

Reagents: Functional programming meets scalable concurrency Aaron Turon Northeastern University Saturday, November 10, 12 Concurrency Parallelism Concurrency is overlapped execution of processes. Parallelism is simultaneous execution of

958 views • 94 slides
Pervasive Parallelism Laboratory Stanford University ppl.stanford.edu Make parallelism

Pervasive Parallelism Laboratory Stanford University ppl.stanford.edu Make parallelism

Kunle Olukotun Pervasive Parallelism Laboratory Stanford University ppl.stanford.edu Make parallelism accessible to all programmers Parallelism is not for the average programmer Too difficult to find parallelism, to debug, maintain

593 views • 40 slides
Microservices at Netflix Scale First Principles, Tradeoffs, Lessons Learned Ruslan Meshenberg

Microservices at Netflix Scale First Principles, Tradeoffs, Lessons Learned Ruslan Meshenberg

Microservices at Netflix Scale First Principles, Tradeoffs, Lessons Learned Ruslan Meshenberg @rusmeshenberg Microservices: all benefits, no costs? Netflix is the worlds leading Internet television network with over 81 million members in

1.01k views • 62 slides
PHI: ARCHITECTURAL SUPPORT FOR SYNCHRONIZATION- AND BANDWIDTH-EFFICIENT COMMUTATIVE SCATTER

PHI: ARCHITECTURAL SUPPORT FOR SYNCHRONIZATION- AND BANDWIDTH-EFFICIENT COMMUTATIVE SCATTER

PHI: ARCHITECTURAL SUPPORT FOR SYNCHRONIZATION- AND BANDWIDTH-EFFICIENT COMMUTATIVE SCATTER UPDATES Anurag Mukkara , Nathan Beckmann, Daniel Sanchez MICRO 2019 Scatter updates are common but inefficient 2 Scatter updates are common but

1.55k views • 152 slides
State of the Art in Microservices Adrian Cockcroft @adrianco Technology Fellow - Battery

State of the Art in Microservices Adrian Cockcroft @adrianco Technology Fellow - Battery

State of the Art in Microservices Adrian Cockcroft @adrianco Technology Fellow - Battery Ventures Microxchg Berlin - February 2015 Speeding up Development Microservice Architectures Whats Next Why am I here? %*&! By Simon Wardley

1.12k views • 98 slides
Roller micro slides type PMMR TECHNISCHE DATEN ASSEMBLY The mounting holes of each type are

Roller micro slides type PMMR TECHNISCHE DATEN ASSEMBLY The mounting holes of each type are

Roller micro slides type PMMR TECHNISCHE DATEN ASSEMBLY The mounting holes of each type are drilled to a standard confjguration in slide-top and -base and permit the user a quick attachment into the application. Thread holes in the table

236 views • 7 slides
A micro-perspective on variation and universals Jeroen van Craenenbroeck 1 Marjo van Koppen 2 1

A micro-perspective on variation and universals Jeroen van Craenenbroeck 1 Marjo van Koppen 2 1

A micro-perspective on variation and universals Jeroen van Craenenbroeck 1 Marjo van Koppen 2 1 CRISSP/KU Leuven, jeroen.vancraenenbroeck@kuleuven.be 2 UiL-OTS/Utrecht University, j.m.vankoppen@uu.nl . . . . . . . . . . . . . . . .

798 views • 45 slides
Towards Automatically Checking Thousands of Failures with Micro-Specifications Haryadi S. Gunawi,

Towards Automatically Checking Thousands of Failures with Micro-Specifications Haryadi S. Gunawi,

Towards Automatically Checking Thousands of Failures with Micro-Specifications Haryadi S. Gunawi, Thanh Do , Pallavi Joshi, Joseph M. Hellerstein, Andrea C. Arpaci-Dusseau , Remzi H. Arpaci-Dusseau , Koushik Sen University of

1.84k views • 98 slides
The Micro-Price Sasha Stoikov Cornell University Jim Gatheral @ NYU Introduction General

The Micro-Price Sasha Stoikov Cornell University Jim Gatheral @ NYU Introduction General

Introduction General Framework Toy models Discrete Markov model Data Analysis Conclusion The Micro-Price Sasha Stoikov Cornell University Jim Gatheral @ NYU Introduction General Framework Toy models Discrete Markov model Data Analysis

639 views • 30 slides
Micro Simulations in ORA Tom Magelinski tmagelin@andrew.cmu.edu School of Computer Science,

Micro Simulations in ORA Tom Magelinski tmagelin@andrew.cmu.edu School of Computer Science,

<Your Name> Micro Simulations in ORA Tom Magelinski tmagelin@andrew.cmu.edu School of Computer Science, Carnegie Mellon Summer Institute 2020 Center for Computational Analysis of Social and Organizational Systems

395 views • 28 slides

More recommend