bounded arbitration algorithm for qos supported on chip
play

Bounded Arbitration Algorithm for QoS-Supported On-chip Com m - PowerPoint PPT Presentation

Bounded Arbitration Algorithm for QoS-Supported On-chip Com m unication Moham m ad Abdullah Al Faruque Gereon Weiss Joerg Henkel Chair for Embedded Systems (CES) University of Karlsruhe (TH) Karlsruhe, Germany Chair for Em bedded System s


  1. Bounded Arbitration Algorithm for QoS-Supported On-chip Com m unication Moham m ad Abdullah Al Faruque Gereon Weiss Joerg Henkel Chair for Embedded Systems (CES) University of Karlsruhe (TH) Karlsruhe, Germany Chair for Em bedded System s 1 / 28 University of Karlsruhe ( TH)

  2. Overview � Motivation � Quality-of-Service (QoS) � Related Works � Bounded Arbitration Algorithm • QoS Specification • Algorithm Description • NoC Simulation Environment � Evaluation & Results � Summary Chair for Em bedded System s 2 / 28 University of Karlsruhe ( TH)

  3. Motivation Next Generation Handheld Devices � The download and the TV continue when an incoming call is accepted Dow nload File X � Games, Sensor nodes, Navigation etc. Phone X TV – Channel … X Huge Computational Power and Application Incoming Video Call! Concurrency � Computational Power � MPSoC � Varying requirements � Implementing QoS mechanism Chair for Em bedded System s 3 / 28 University of Karlsruhe ( TH)

  4. Motivation Com m unication centric ( NoC based) MPSoC Bus based MPSoC design system bus � Packet based communication � Computation Power is provided ARB � Highly Scalable by multiple processing units • Quality of Service needs to be MEM � QoS is implemented by implemented in the packet μ P μ P different types of buses μ P • Needs to differentiate at transaction level μ P hi-perf FPGA bus ARB B • Needs to provide tight guarantee DSP MPEG (i.e. guaranteed BW) � Problems with Bus communication UART • Resource utilization needs to be μ P DSP • Deep Sub Micron problems kept minimum ARB B • Scalability issues Efficient QoS Mechanism at a ETH [ M. Horowitz et. al.] peripheral m inim um cost is necessary bus Chair for Em bedded System s 4 / 28 University of Karlsruhe ( TH)

  5. Quality Of Services Guarantees / Probabilistic on � Perform ance related guarantee � Reliability related guarantee Perform ance : � Max end-to-end latency � Min throughput (% Bandwidth ) � Max deviation of latency ( Jitter ) Reliability : � In-order data transmission � Correctness of data � No loss of data (Lossless transmission) � Availability Service Class Specification Chair for Em bedded System s 5 / 28 University of Karlsruhe ( TH)

  6. Overview � Motivation � Quality-of-Service (QoS) � Related W orks � Bounded Arbitration Algorithm � Evaluation & Results � Summary Chair for Em bedded System s 6 / 28 University of Karlsruhe ( TH)

  7. Related W orks � Connection Based Approach: i.e. Æthereal • K. Goossens et al. Æthereal Network on Chip: Concepts, Architectures, and Implementations, 2005. • E. Rijpkema et al. Trade-offs in the design of a router with both guaranteed and best-effort services for networks on chip , 2003. � Service Class Based Approach: i.e. DiffServ, QNoC • Evgeny Bolotin et al. QNoC: QoS architecture and design process for network on chip , 2004. • M. D. Harmanci et al. Quantitative modeling and comparison of communication schemes to guarantee quality-of-service in networks-on-chip , 2005. • N. Kavaldjiev et al. A virtual channel Network-on-Chip for GT and BE traffic , 2006. Chair for Em bedded System s 7 / 28 University of Karlsruhe ( TH)

  8. Related W orks Connection Based Service Class Based Advantages + TDMA / Guaranteed throughput + High resource utilization Contention free routing + + Priority aware service Buffer reduction + Disadvantages - Connection management - No connections (Relative guarantees) - Fixed resource reservation Contention / Starvation - Classification at design time Buffer requirements - - Lookup-tables Inflexibility of classification - - Underutilization No Hard Guarantees Chair for Em bedded System s 8 / 28 University of Karlsruhe ( TH)

  9. Focus of W ork High Utilization & Tight Guarantees Possible solution can be: � Adaptive approach with sharing of resources • Buffers: Not assigned to particular class • Links: With efficient arbitration Our Link Arbitration Bounded Arbitration Algorithm ( BAA) Chair for Em bedded System s 9 / 28 University of Karlsruhe ( TH)

  10. Overview � Motivation � Quality-of-Service (QoS) � Related Works � Bounded Arbitration Algorithm • QoS Specification • Algorithm Description • NoC Simulation Environment � Evaluation & Results � Summary Chair for Em bedded System s 10 / 28 University of Karlsruhe ( TH)

  11. QoS Specification Service Class Latency Bandw idth Jitter Exam ple Min./ Max. Priority H Low … … Signal Priority X 1 … Fixed No / Video Stream Guarantee Fixed … … … … … Priority X n … Fixed Data ok Stream Guarantee Priority BE As low as Best Not fixed ok possible Effort Chair for Em bedded System s 11 / 28 University of Karlsruhe ( TH)

  12. Bounded Arbitration Algorithm Slot_assigned = FALSE Slot_assigned = FALSE Initialize Initialize !slot_assigne !slot_assigne d d Yes Yes k <= k <= VC jitter VC delay_jitter k k Yes Yes slotk < C p max slotk < C p max & ST avl No & ST avl No Yes Yes No Assign the No No slots Assign the No slots First Part Middle Part Last Part statements statements • Highest Priority transctions • • Rest of the Best Effort The Scheduling Table is Initialize get through first with lower now filled with latency traffics are now allowed till Initialize bound slot allocation their upper bound assigned sensitive transactions slot Chair for Em bedded System s 12 / 28 University of Karlsruhe ( TH)

  13. Bounded Arbitration Algorithm Classes Min / Max BW Latency Jitter Priority 3: 30% / 50% yes yes Priority 2: 30% / 30% yes no Priority 1: 10% / 100% no yes Service Class VC 3 ARBITER Service 3 3 3 2 2 2 1 3 3 1 1 Class VC 2 Link Service Class VC 1 Chair for Em bedded System s 13 / 28 University of Karlsruhe ( TH)

  14. NoC Sim ulation Environm ent I nput Flow Control Port Area And other Area + Const Output memory Port Error Control Area And Block Control-bits Memory Req Area and no# err arbiter Area + Bits Const memory Positional Virtual Channel Crossbar Router Design Area and buffer Area & power depth Const Area consumption Link Network Design I nterface Length/ Area Area Synthesizable NoC Synthesizable ASI Ps � SystemC model for VHDL Model for Synthesis Simulation Hardw are Evaluation ( ProDesign FPGA Board) Chair for Em bedded System s 14 / 28 University of Karlsruhe ( TH)

  15. NoC Sim ulation Environm ent VCS: Virtual Channel Selector I D: Input Decoder VC: Virtual Channel OA : Output Arbiter OD: Output Decoder VCA: Virtual channel Arbiter PCA: Physical Channel Arbiter Adm ission Buffer Packet control resources control � Cycle-accurate design � Deeply pipelined structure Chair for Em bedded System s 15 / 28 University of Karlsruhe ( TH)

  16. Overview � Motivation � Quality-of-Service (QoS) � Related Works � Bounded Arbitration Algorithm � Evaluation & Results � Summary Chair for Em bedded System s 16 / 28 University of Karlsruhe ( TH)

  17. Case study: MPEG4 Video Decoder MPEG4 Video Decoder • Are the guarantees met? • Comparison: BAA, RR and FAA - BAA: Bounded Arbitration Algorithm - RR: Round Robin Arbitration - FAA: Fixed Arbitration Algorithm • Latency, Throughput ! • Waste of bandwidth ! • Granularity of specification ! Chair for Em bedded System s 17 / 28 University of Karlsruhe ( TH)

  18. Traffic Specifications Service Min Max Latency Jitter Class allow ed BW BW Sensitive MPEG4 800 MB/ s 2000 MB/ s yes yes Stim ulus1 400 MB/ s 2000 MB/ s no yes Stim ulus2 400 MB/ s 2000 MB/ s no yes Stim ulus3 400 MB/ s 2000 MB/ s no yes Chair for Em bedded System s 18 / 28 University of Karlsruhe ( TH)

  19. MPEG4 Mapping Stimulus3 IP IP IP IP IP Producer (0x2) NI NI NI NI NI S S S S S Stimulus1 Stimulus1 IP IP IP IP iDCT INPUT iQUANT IP Producer Consumer (1x1) (1x3) (1x2) (1x0) (1x4) NI NI NI NI NI S S S S S Stimulus2 Stimulus2 IP IP IP IP IP IO PREDICT ADD Consumer Producer (2x3) (2x2) (2x3) (2x4) (2x0) NI NI NI NI NI S S S S S Stimulus3 IP IP IP IP IP Consumer (3x2) NI NI NI NI NI S S S S S Chair for Em bedded System s 19 / 28 University of Karlsruhe ( TH)

  20. Bandw idth Sharing Round Robin Arbitration Bandwidth Arbitration Stim ulusB Stim ulusC Stim ulusA 5 0 0 MB/ s 5 0 0 MB/ s 5 0 0 MB/ s 5 0 0 MB/ s MPEG4 Min BW requirem ent for MPEG4 8 0 0 MB/ s Chair for Em bedded System s 20 / 28 University of Karlsruhe ( TH)

  21. Bandw idth Sharing Bounded Arbitration Algorithm Bandwidth Arbitration Stim ulusB Stim ulusC Stim ulusA 4 0 0 MB/ s 4 0 0 MB/ s 4 0 0 MB/ s 8 0 0 MB/ s MPEG4 Min BW requirem ent for MPEG4 8 0 0 MB/ s Chair for Em bedded System s 21 / 28 University of Karlsruhe ( TH)

  22. Throughput Com parison Average throughput of all connections 2500 Throughput (MB/s) 2000 1500 1000 500 0 MC Inp iQuant Inp Inp iDCT S1 S2 S3 Connections iQuant iDCT MC iDCT ADD ADD Stim ulus1 -3 MPEG4 BAA RR FAA Chair for Em bedded System s 22 / 28 University of Karlsruhe ( TH)

Recommend


More recommend