1 / 21

CS 8501 Networks-on-Chip ( NoCs )

CS 8501 Networks-on-Chip ( NoCs ). Lukasz Szafaryn 15 FEB 10. Motivation. Bus has been the most popular interconnect for multiprocessor systems When scaling feature sizes and frequency, wire delays remain larger than clock cycle.

rupert
Download Presentation

CS 8501 Networks-on-Chip ( NoCs )

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 8501Networks-on-Chip (NoCs) Lukasz Szafaryn 15 FEB 10

  2. Motivation • Bus has been the most popular interconnect for multiprocessor systems • When scaling feature sizes and frequency, wire delays remain larger than clock cycle • When expanding to a many core-system, contention decreases throughput • Need for interconnect with deterministic delays and scalability

  3. What is a Network-on-Chip (NoC)? • Leveraging existing computer networking principles to improve inter-component intra-chip communications • Each on-chip component connected by an intelligent switch to particular communication wire(s) • Improvement over standard bus based interconnections for SoC architectures in terms of throughput

  4. Topologies CLICHÉ Torus Folded torus

  5. Topologies contd. SPIN Octagon BFT

  6. Switching Circuit Switching • Dedicated path, or circuit, is established over which data packets will travel • Naturally lends itself to time-sensitive guaranteed service due to resource allocation • Reservation of bandwidth decreases overall throughput and increases average delays Packet Switching • Intermediate routers are now responsible for the routing of individual packets through the network, rather than following a single path • Provides for so-called best-effort services • Sharing of resources allows for higher throughput

  7. Switching contd. Wormhole Switching • Message is divided up into smaller, fixed length flow units called flits • Only first flit contains routing information, subsequent flits follow • Buffer size is significantly reduced due to the limitation on the number of flits needed to be buffered at any given time Virtual Channels • Allows for several instances of wormhole switching • Additional buffers are added, which increases overall switch size, but significantly increases throughput

  8. Performance Metrics Simulator developed to measure: • Throughput (in flits) • Latency (of flits) • Energy (per packet) Hardware model developed to estimate: • Area (router and link overhead)

  9. Simulator Setup • SoC with 256 fixed-size (100K gate) elements • Topologies use the same type of switching (wormhole with usually 4 virtual channels) • Topologies use different routing schemes (different configurations of switches and links)

  10. Simulated Traffic • Uniform and localized traffic patterns • Poisson and self-similar injection methods

  11. Number of Virtual Channels Throughput Latency

  12. Number of Virtual Channels contd. Energy Dissipation

  13. Injection Load Accepted Traffic Latency

  14. Injection Load contd. Energy Dissipation

  15. Localization Throughput

  16. Localization contd. Latency (30% local traffic) Latency (80% local traffic)

  17. Localization contd. Energy Dissipation (30% local traffic) Energy Dissipation (80% local traffic)

  18. Interconnect Area Die size: 20mm x 20mm

  19. Interconnect Area contd.

  20. Case Study

  21. Conclusions • NoC is an interconnect architecture of choice of multiprocessor SoC and many-core systems • The main trade-off is between performance and energy • Other considerations could lead to different energy/performance numbers

More Related