HiRA: A Methodology for Deadlock Free Routing in Hierarchical Networks on Chip

Similar documents
HiRA: A Methodology for Deadlock Free Routing in Hierarchical Networks on Chip

Bandwidth Aware Routing Algorithms for Networks-on-Chip

Deadlock and Livelock. Maurizio Palesi

A Thermal-aware Application specific Routing Algorithm for Network-on-chip Design

NOC Deadlock and Livelock

JUNCTION BASED ROUTING: A NOVEL TECHNIQUE FOR LARGE NETWORK ON CHIP PLATFORMS

Extended Junction Based Source Routing Technique for Large Mesh Topology Network on Chip Platforms

Noxim the NoC Simulator

Sanaz Azampanah Ahmad Khademzadeh Nader Bagherzadeh Majid Janidarmian Reza Shojaee

Fault-adaptive routing

PDA-HyPAR: Path-Diversity-Aware Hybrid Planar Adaptive Routing Algorithm for 3D NoCs

Fault-Tolerant Multiple Task Migration in Mesh NoC s over virtual Point-to-Point connections

Lecture 24: Interconnection Networks. Topics: topologies, routing, deadlocks, flow control

Interconnection Networks: Routing. Prof. Natalie Enright Jerger

On Packet Switched Networks for On-Chip Communication

EC 513 Computer Architecture

Bandwidth-aware routing algorithms for networks-on-chip platforms M. Palesi 1 S. Kumar 2 V. Catania 1

Synchronized Progress in Interconnection Networks (SPIN) : A new theory for deadlock freedom

Deadlock. Reading. Ensuring Packet Delivery. Overview: The Problem

A Novel NoC Routing Design Methodology Based on Divide-Conquer Approach *

Lecture 12: Interconnection Networks. Topics: dimension/arity, routing, deadlock, flow control

SYSARC 768 No. of Pages 14, Model 5+ ARTICLE IN PRESS UNCORRECTED PROOF

Deadlock-free XY-YX router for on-chip interconnection network

Deadlock Free Routing in Mesh Networks on Chip with Regions

Comparison of Deadlock Recovery and Avoidance Mechanisms to Approach Message Dependent Deadlocks in on-chip Networks

Generalized Theory for Deadlock-Free Adaptive Wormhole Routing and its Application to Disha Concurrent

Generic Methodologies for Deadlock-Free Routing

4. Networks. in parallel computers. Advances in Computer Architecture

Lecture 15: PCM, Networks. Today: PCM wrap-up, projects discussion, on-chip networks background

Performance Analysis of NoC Architectures

Topologies. Maurizio Palesi. Maurizio Palesi 1

Switching/Flow Control Overview. Interconnection Networks: Flow Control and Microarchitecture. Packets. Switching.

Lecture: Interconnection Networks. Topics: TM wrap-up, routing, deadlock, flow control, virtual channels

Demand Based Routing in Network-on-Chip(NoC)

Towards A Formally Verified Network-on-Chip

Topologies. Maurizio Palesi. Maurizio Palesi 1

Performance Evaluation of Different Routing Algorithms in Network on Chip

Recall: The Routing problem: Local decisions. Recall: Multidimensional Meshes and Tori. Properties of Routing Algorithms

BARP-A Dynamic Routing Protocol for Balanced Distribution of Traffic in NoCs

A Novel Semi-Adaptive Routing Algorithm for Delay Reduction in Networks on Chip

Lecture 7: Flow Control - I

A Reconfigurable Crossbar Switch with Adaptive Bandwidth Control for Networks-on

Lecture 12: Interconnection Networks. Topics: communication latency, centralized and decentralized switches, routing, deadlocks (Appendix E)

Lecture 16: On-Chip Networks. Topics: Cache networks, NoC basics

in Oblivious Routing

Routing Algorithms, Process Model for Quality of Services (QoS) and Architectures for Two-Dimensional 4 4 Mesh Topology Network-on-Chip

TDT Appendix E Interconnection Networks

High Performance Interconnect and NoC Router Design

EECS 578 Interconnect Mini-project

Università degli Studi di Catania Dipartimento di Ingegneria Informatica e delle Telecomunicazioni (DIIT) NoC Research Summary

Trading hardware overhead for communication performance in mesh-type topologies

ReNoC: A Network-on-Chip Architecture with Reconfigurable Topology

Heuristics Core Mapping in On-Chip Networks for Parallel Stream-Based Applications

Evaluation of NOC Using Tightly Coupled Router Architecture

SoC Design. Prof. Dr. Christophe Bobda Institut für Informatik Lehrstuhl für Technische Informatik

A Deterministic Fault-Tolerant and Deadlock-Free Routing Protocol in 2-D Meshes Based on Odd-Even Turn Model

Interconnection Networks: Topology. Prof. Natalie Enright Jerger

Static Virtual Channel Allocation in Oblivious Routing

Network on Chip Architecture: An Overview

CONGESTION AWARE ADAPTIVE ROUTING FOR NETWORK-ON-CHIP COMMUNICATION. Stephen Chui Bachelor of Engineering Ryerson University, 2012.

Low Latency and Energy Efficient Multicasting Schemes for 3D NoC-based SoCs

AS THE NUMBER of cores integrated into a systemon-chip

Dynamic Stress Wormhole Routing for Spidergon NoC with effective fault tolerance and load distribution

Multi-path Routing for Mesh/Torus-Based NoCs

A NEW DEADLOCK-FREE FAULT-TOLERANT ROUTING ALGORITHM FOR NOC INTERCONNECTIONS

FT-Z-OE: A Fault Tolerant and Low Overhead Routing Algorithm on TSV-based 3D Network on Chip Links

NOW Handout Page 1. Outline. Networks: Routing and Design. Routing. Routing Mechanism. Routing Mechanism (cont) Properties of Routing Algorithms

BISTed cores and Test Time Minimization in NOC-based Systems

Efficient And Advance Routing Logic For Network On Chip

IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 22, NO. 1, JANUARY

Lecture 3: Flow-Control

Escape Path based Irregular Network-on-chip Simulation Framework

DESIGN AND IMPLEMENTATION ARCHITECTURE FOR RELIABLE ROUTER RKT SWITCH IN NOC

EECS 570. Lecture 19 Interconnects: Flow Control. Winter 2018 Subhankar Pal

Dynamic Routing of Hierarchical On Chip Network Traffic

Lecture 13: Interconnection Networks. Topics: lots of background, recent innovations for power and performance

SoC Design Lecture 13: NoC (Network-on-Chip) Department of Computer Engineering Sharif University of Technology

Thomas Moscibroda Microsoft Research. Onur Mutlu CMU

Noc Evolution and Performance Optimization by Addition of Long Range Links: A Survey. By Naveen Choudhary & Vaishali Maheshwari

Performance Oriented Docket-NoC (Dt-NoC) Scheme for Fast Communication in NoC

Whole Packet Forwarding: Efficient Design of Fully Adaptive Routing Algorithms for Networks-on-Chip

Global Adaptive Routing Algorithm Without Additional Congestion Propagation Network

SURVEY ON LOW-LATENCY AND LOW-POWER SCHEMES FOR ON-CHIP NETWORKS

ERA: An Efficient Routing Algorithm for Power, Throughput and Latency in Network-on-Chips

Bursty Communication Performance Analysis of Network-on-Chip with Diverse Traffic Permutations

Performance Evaluation of Mesh with Source Routing for Packet Loss

Basic Switch Organization

Flow Control can be viewed as a problem of

ECE 4750 Computer Architecture, Fall 2017 T06 Fundamental Network Concepts

Akash Raut* et al ISSN: [IJESAT] [International Journal of Engineering Science & Advanced Technology] Volume-6, Issue-3,

Design and Test Solutions for Networks-on-Chip. Jin-Ho Ahn Hoseo University

ECE 669 Parallel Computer Architecture

Communication Performance in Network-on-Chips

MESH-CONNECTED networks have been widely used in

EE 382C Interconnection Networks

AC : HOT SPOT MINIMIZATION OF NOC USING ANT-NET DYNAMIC ROUTING ALGORITHM

WITH the development of the semiconductor technology,

A Dynamic NOC Arbitration Technique using Combination of VCT and XY Routing

A Hybrid Interconnection Network for Integrated Communication Services

Mapping of Real-time Applications on

Transcription:

HiRA: A Methodology for Deadlock Free Routing in Hierarchical Networks on Chip Rickard Holsmark 1, Maurizio Palesi 2, Shashi Kumar 1 and Andres Mejia 3 1 Jönköping University, Sweden 2 University of Catania, Italy 3 Technical University of Valencia, Spain NoC Symposium San Diego, May 10 13, 2009

Contents Introduction Hierarchical networks Deadlock free hierarchical routing Hierarchical deadlock free routing methodology Basic idea and theoretical framework Design methodology Evaluation Simulation results Conclusions 2

The Road to Network on Chip and Beyond Transistor Integrated Circuits Microprocessor 1960 1970 1980 System on Chip 1990 Networks on Chip (NoC) Networks of NoCs Evolution of building blocks in electronic system design 2000 2010 3

Hierarchical Networks Hierarchical networks in system design Reuse of networks as subnets Partitioning large networks Examples of hierarchical NoCs Hybrid ring/mesh Optimization global/local traffic (Bourduas and Zilic, 2007) HiNoC, mesh/fat-tree Efficient GALS, asynchronous mesh/synchronous tree (Hollstein et al., 2006) Our approach to hierarchical networks Heterogeneous interconnect Subnets and subnet interconnect Routing Use optimal routing algorithm for each subnet 4

Issues when Combining Networks Different structures Switching Store and forward wormhole Routing control Source distributed Deadlock strategy Avoidance recovery Physical implementation Given identical structures Different routing algorithms -> deadlock problems 5

Deadlock Free Routing when Combining Networks Each subnet deadlock free Deadlock free? -Whole network must be considered Routing Alg. S1 Routing Alg. S2 Shared Mem Routing Alg. N Shared Mem Network Routing Alg. S3 Routing Alg. S4 6

Hierarchical Deadlock Free Routing Each subnet deadlock free Deadlock free? Deadlock free subnets + rules + external routing algorithm Routing Alg. S1 Routing Alg. S2 Routing Alg. H Shared Mem Routing Alg. S1 Routing Alg. S2 Shared Mem Network Routing Alg. E Routing Alg. S3 Routing Alg. S4 Routing Alg. S3 Routing Alg. S4 7

Revising Packet Deadlocks Deadlock Packets involved in a circular wait Risk increased by using wormhole switching Avoid/recover/drop Avoidance techniques Analysis: Channel dependency graphs (CDG), Turn-model, Channel wait for graphs (CWG) Implementation: Restrict packet routes Examples of algorithms Odd-even, X-Y, Negative-first, Up/Down 8

Deadlock Example Four packets: P1,..,P4 Require turns:,.., -Based on previous decisions (turns) Packet Deadlock No Packet Deadlock P1 Restrict turn 1 2 1 2 P2 P2 P1 P4 P4 3 4 3 4 P3 P3 9

Deadlock Freedom and CDG Topology 1 2 Channel dependency graph (CDG) l 21 3 4 l 13 l 31 l 12 l 43 l 24 l 42 Cycles in CDG = not deadlock free l 34 Cut all cycles Implement routing restrictions l 21 l 12 1 2 l 13 l 31 l 24 l 42 = deadlock free l 43 3 4 l 34 10

Deadlocks when Interconnecting Networks Two deadlock free networks interconnected Subnet A Subnet B Possibility of cycle in CDG May not be deadlock free and connected Analyze CDG Bi-directional routing restriction Cycle 11

Deadlock Free Interconnection of Networks The two networks again No CDG cycle formed Deadlock free and connected Safe interconnection nodes Subnet A Subnet B Analyze CDG Moved routing restrictions No cycle 12

Deadlock Free Hierarchical Routing Method for interconnecting deadlock free (sub-) networks Resulting network is deadlock free and connected Internal subnet routing algorithms un-changed Hierarchical network structure Subnets Networks with independent (deadlock free) routing algorithms Boundary nodes Subset of nodes in a subnet that connects to other subnets External links Links between subnets (are not in any subnet) 13

Hierarchical Network Structure Subnet S 1 S 3 Internal nodes/links S 2 External link S 4 S 6 Boundary node S 5 14

Hierarchical Routing Algorithm Building blocks Subnet routing algorithms Given, routes inside subnets (internal subnet links) External routing algorithm Routes between subnets (external and internal links) Safe boundary nodes (interconnections nodes) Prohibits un resolvable CDG cycles Defined in each subnet Complete hierarchical routing algorithm Combination of subnet internal and external algorithms Usage of safe boundary nodes enable deadlock freedom 15

Hierarchical Route Example Routing a packet from subnet S i to subnet S j o Source node n s, destination node n d 1. Node n s : R G (n d )=R i (b s ) (internal route) 2. Node b s : R G (n d )=R G (b d ) (external route) b s Internal routing Subnet Salgorithm R i i Internal routing restricitons n s Internal routing algorithm R i 3. Node b d : R G (n d )=R j (n d ) (internal route) b d n d External routing algorithm R G External routing restriciton Subnet S j 16

Safe Boundary Nodes Example of safe boundary nodes in a subnet Routing algorithm in subnet S 6 : Negative first Subnet S 6 CDG Negative-first routing restrictions Safe if no CDG path All safe Boundary nodes (4 of 6) 17

Theoretical Foundation Theorem: A hierarchical routing algorithm is deadlock free and connected if all subnet routing algorithms are deadlock free, the external routing algorithm is deadlock free and subnets are interconnected by safe boundary nodes Proof logic Subnet CDGs and external CDG acyclic External CDG: Restrictions on safe boundary nodes Combined dependencies Safe boundary nodes: Only from one boundary node to another boundary node Acyclic CDG for combined (hierarchical) algorithm: Deadlock free All restricted external routes included in analysis: Connected 18

Identification of Safe Boundary Nodes Termination Not succesful termination Succesful termination Unsafe Cycle enabled Safe Cycle disabled l 13 l 21 l 12 l 31 l 43 l 24 l 42 l 13 l 21 l 12 l 31 l 43 l 24 l 42 l 34 l34 19

Finding Safe Nodes Possible Safe nodes? Depends on routing algorithm Two main types (Algorithms with Deadlock freedom based on acyclic CDG) Deterministic routing All network nodes are safe nodes (assuming bi-directional links) Adaptive routing Not all nodes are safe nodes Number of and positions of safe nodes vary 20

Designing External Routing Algorithm Deadlock free external routing algorithm Un-changed internal algorithms Internal paths between safe boundary nodes given If cycles in external CDG: Restrictions on boundary node links (external internal, external external) Affects only external traffic External routing algorithm = normal routing algorithm Links: External links and internal paths between boundary nodes Nodes: Boundary nodes and internal nodes on internal boundary node paths 21

Example: External Algorithm External routing restrictions Cycle free external CDG (effectíve dependencies, safe nodes) S 1 S 3 S 4 S 2 X-Y S 6 S 5 External restrictions 22

Interconnection of Y X and X Y Networks Y-X X-Y Y-X Y-X X-Y Y-X Simply connect: -Cycle, not deadlock free Y-X External routing X-Y X-Y Y-X Hierarchical: -No cycles, deadlock free No path Not allowed (X-Y) 23

Evaluation Set Up Noxim network simulator Wormhole switching Packet size randomly between 2 and 8 flits Traffic types Synthetic: Standard types (random, transpose, ) Application: Communication graphs from research papers (Hu and Marculescu, 2005), (Yoon, Lee and Park, 2006), (Jaspers and de With, 1999), (van der Tol and Jaspers, 2001) Destinations Local traffic: Communication within subnets Global traffic: Communication between subnets 24

Average Delay Synthetic Traffic Average delay for each routing algorithm under varying global /local ratio 100 90 80 70 Four subnets with different traffic types Average delay (cycles) 60 50 40 30 XY Odd-Even West-First Hierarchical 20 10 0 0% 2% 4% 9% 13% 17% 22% Global to Local ratio 25

Throughput Jitter Application Traffic Normalized throughput jitter for each algorithm and application Five core subnets + one memory subnet Normalized Throughput Jitter 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00 NegativeFirst WestFirst APSRA XY NorthLast OddEven Hierarchical MMS-Enc MMS-Dec PIP MWD MIMO-OFDM 26

Conclusions Concept of hierarchy is useful for designing large NoCs Helps to raise the level of reuse from components to subnets Building networks with heterogeneous topology, heterogeneous routing,... Partition a large network for mapping applications, testing, fault tolerance,... New Issues and concepts Deadlock free routing and connectivity Safe boundary nodes Challenges and future work Pessimistic assumption of interconnecting subnets using only safe boundary nodes Unsafe nodes can be used in certain cases Node addressing and router design Traffic management and congestion control in hierarchical networks 27

Awake? Thank you for Listening! Questions? 28