trade offs in the design of a router with both guaranteed
play

Trade Offs in the Design of a Router with Both Guaranteed and - PowerPoint PPT Presentation

Trade Offs in the Design of a Router with Both Guaranteed and Best-Effort Services for Networks on Chip E. Rijpkema, K. Goossens, A. R dulescu, J. Dielissen, J. van Meerbergen, P. Wielage, and E. Waterlander why Networks-on-Chip problems


  1. Trade Offs in the Design of a Router with Both Guaranteed and Best-Effort Services for Networks on Chip E. Rijpkema, K. Goossens, A. R dulescu, J. Dielissen, J. van Meerbergen, P. Wielage, and E. Waterlander

  2. why Networks-on-Chip problems observed for SoC design • deep sub micron • design complexity • wire cost • increasing # of IP blocks • timing closure • increasing dynamism decouple computation from communication application demands application network IP IP IP independent presentation session services transport network network R R R data link dependent physical router network hardware technology Æthereal Network-on-Chip 2 Philips Research

  3. why Networks-on-Chip • problems observed for SoC design • deep sub micron • design complexity • key: decouple computation from communication application demands application network IP IP IP independent presentation session services transport network network R R R data link dependent physical router network hardware technology Æthereal Network-on-Chip 3 Philips Research

  4. outline • services • combined router architecture • guaranteed throughput router architecture • best-effort router architecture • router prototype • conclusions Æthereal Network-on-Chip 4 Philips Research

  5. services I • we need a network that is • predictable • cost effective application demands requirements guarantees services efficiency constraints hardware technology • build guarantees on top of guarantees • efficient network is efficient at every layer Æthereal Network-on-Chip 5 Philips Research

  6. services II • timeless guarantees • guaranteed data integrity best-effort • guaranteed data delivery service (BE) • guaranteed in-order delivery guaranteed throughput service • time related guarantees (GT) (over bounded time interval) • guaranteed throughput • guaranteed latency Æthereal Network-on-Chip 6 Philips Research

  7. guarantees vs. best-effort • GT requires dimensioning for guaranteed throughput • BE requires dimensioning for average throughput r wc r wc r avg 1 2 3 4 1 2 3 4 1 2 3 4 time time time guaranteed guaranteed guaranteed combination guaranteed delivery throughput delivery is beneficial throughput (bounded interval) Æthereal Network-on-Chip 7 Philips Research

  8. BE & GT combined architecture • conceptually, two disjoint routers • a router with GT service class • a router with BE service class r wc BE router programming 1 2 3 4 GT router priority/arbitration • to obtain an efficient combination routers must have similar architectures Æthereal Network-on-Chip 8 Philips Research

  9. buffering strategy 1 1 • output queuing 2 2 • highest cost � � � � • highest performance N N 1 1 • input queuing 2 2 X • lowest cost � � � • lowest performance N N 1 1 • virtual output queuing 2 2 • moderate cost X � � � • high performance N N preferred solution Æthereal Network-on-Chip 9 Philips Research

  10. contention • links in network are shared resources • contention occurs when multiple data request same link at same time • GT and BE resolve contention differently Æthereal Network-on-Chip 10 Philips Research

  11. guaranteed throughput • to guarantee latency or bandwidth over finite interval • cannot drop data • must bound contention • rate-based scheduling • has high buffer costs (deep fifos/output queuing) • deadline-based scheduling • even higher buffer costs (deep priority queues) • contention-free routing • low buffer costs (shallow fifos) Æthereal Network-on-Chip 11 Philips Research

  12. contention-free routing I • scheduling packet injection in network to avoid contention • in space: disjoint paths as in pure circuit switching • in time: time-division multiplexing as with a statically scheduled bus • in time and space: our solution Æthereal Network-on-Chip 12 Philips Research

  13. contention-free routing II • divide time in slots time block 1 block 2 block 3 slot • a block is amount of data that fits in a slot • block entering router in slot n enters next in slot n+1 n n+1 n+3 n+2 1 1 2 2 • matches with input queuing X � � � N N Æthereal Network-on-Chip 13 Philips Research

  14. contention-free routing III • routers have tables that • store contention resolution & routing information • allow distributed programming 0 0 1 0 - 0 - - 0 0 0 0 0 1 1 - - 1 1 1 1 1 1 - 1 1 0 - 2 2 2 2 2 - - 1 1 1 3 3 3 3 3 - - - - - 4 4 4 4 4 - - - - - S-1 S-1 S-1 S-1 S-1 • small blocks � low buffering cost � low latency small slots � throughput guarantee on smaller period Æthereal Network-on-Chip 14 Philips Research

  15. best-effort architecture • to ensure high resource utilization • statistical multiplexing • packet-switching • but implement BE service class • packet-switching • network flow control (routing mode) • contention resolution Æthereal Network-on-Chip 15 Philips Research

  16. packets and flits • packet = header + payload payload H • packet might be transmitted in smaller parts called flits flit 1 flit 2 flit 3 flit 4 • flits divide time in iterations and must be scheduled time flit 1 flit 2 flit 3 flit 4 • smaller flit size � higher scheduling rate � lower latency � less storage Æthereal Network-on-Chip 16 Philips Research

  17. network flow control (routing mode) performance/cost per router latency storage network flow control • store and forward routing packet packet • first receive whole packet • then transmit whole packet • virtual cut-through routing flit packet • send flit immediately • if next router can receive entire packet • wormhole routing flit flit • send flit immediately • if next router can receive that flit Æthereal Network-on-Chip 17 Philips Research

  18. contention resolution • queuing at input � set paths from inputs to outputs • router has switch • bipartite graph matching 1 1 X 2 2 3 3 • algorithm must • be fair • have low complexity (to schedule at flit rate) • approximation of maximal matching Æthereal Network-on-Chip 18 Philips Research

  19. combining GT and BE • links must be shared by GT and BE traffic • grain size of interleaving must match • block size = flit size • smallest value for this is given by implementation • minimize scheduler latency � L • maximize data path speed � F • flit size = block size = F · L Æthereal Network-on-Chip 19 Philips Research

  20. router prototype • snapshot of current control prototype router: • input queuing • arity 5 • 32 bits wide words • 8 flits deep BE queues • 256 slots • 0.25 mm 2 CMOS12 • 500 MHz data path • 166 MHz control path • flit size is 3 words • throughput per link: 500 MHz ·32 bits = 16Gb/s Æthereal Network-on-Chip 20 Philips Research

  21. conclusions • for NoCs, guaranteed services are essential • demonstrated the useful combination of: • BE service class � timeless guarantees • GT service class � BE + time related guarantees • made trade-offs to come to efficient combined router • proved feasibility with router prototype Æthereal Network-on-Chip 21 Philips Research

  22. router prototype • snapshot of current prototype router: • 5 input and 5 output ports (arity 5) • 0.25 mm 2 CMOS12 • 500 MHz data path, 166 MHz control path • flit size of 3 words of 32 bits • 500x32 = 16 Gb/s throughput per link • 256 slots & 5x1 flit fifos for guaranteed-throughput traffic • 6x8 flit fifos for best-effort traffic Æthereal Network-on-Chip 22 Philips Research

  23. control control Æthereal Network-on-Chip 23 Philips Research

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


More recommend