1 / 8

Performance Measures and Interconnection Networks: Fundamentals of Parallel Software Design

This lecture focuses on the essential concepts in performance measures and interconnection networks critical for parallel software design. Key topics include cache coherence mechanisms such as snooping and directory-based approaches, emphasizing how to manage data consistency in multi-core systems. It covers measures like speedup, efficiency, and Amdahl's law, while presenting practical problems involving network architectures like crossbars and hypercubes. Understanding thread safety, critical sections, and race conditions is also crucial for designing effective parallel programs.

uriel-oneil
Télécharger la présentation

Performance Measures and Interconnection Networks: Fundamentals of Parallel Software Design

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. CS 470/570 Lecture 4 • Performance measures • Interconnection networks • Cache coherence • Parallel Software • Parallel Program design • Example Problems

  2. Performance Measures • Speedup • Tserial/Tparallel • Efficiency • Speedup/number of processors • Amdahl’s law • Scalable

  3. Interconnection Networks • Bus • Crossbar • Fully connected • Hypercube • Omega • Latency • Bandwidth

  4. Cache Coherence • Snooping cache coherence • Directory based cache coherence

  5. Parallel Software • Nondeterminism • Race conditions • Critical section • Mutual exclusion • Thread Safe

  6. Parallel Program Design • Partitioning • Communication • Aggregation

  7. Example Problems • 1. Draw a 4X4 crossbar switch. • 2. Draw a 4X4 Omega network • 3. Suppose there are 16 processors connected in a 4 dimensional hypercube. • a. How many direct connections to other processors does each processor have? • b. What is the minimum number of links that will have to be traversed to send data from processor 3 to processor 12?

  8. Example Problems • 4. Suppose you have a machine with 4 cores that uses snooping cache coherence. Core 0 reads from address 10,000, core 3 reads from address 10,000. Core 3 writes to address 10,000. How does Core 0 know its copy of the contents of address 10,000 is out of date. • 5. On a 4X4 Omega network can the following source destination requests be satisfied at the same time. (0,0), (3,1), (1,2), (2,3) • 6. How many switches are in an 8X8 crossbar? • 7. How many switches are on an 8X8 Omega network?

More Related