NoC Impact on Design Methodology. Characteristics of NoC-based design. Timing Closure in Traditional VLSI

Similar documents
World Academy of Science, Engineering and Technology International Journal of Civil and Environmental Engineering Vol:4, No:10, 2010

Exercise 8. Panel Data (Answers) (the values of any variable are correlated over time for the same individuals)

APPLYING BI-OBJECTIVE SHORTEST PATH METHODS TO MODEL CYCLE ROUTE-CHOICE ABSTRACT

Time & Distance SAKSHI If an object travels the same distance (D) with two different speeds S 1 taking different times t 1

DIFFUSION ESTIMATION OF MIXTURE MODELS WITH LOCAL AND GLOBAL PARAMETERS

A study of volatility risk

THE PERFORMANCE OF ALTERNATIVE INTEREST RATE RISK MEASURES AND IMMUNIZATION STRATEGIES UNDER A HEATH-JARROW-MORTON FRAMEWORK

Interval Type-1 Non-Singleton Type-2 TSK Fuzzy Logic Systems Using the Kalman Filter - Back Propagation Hybrid Learning Mechanism

Kinematics. Overview. Forward Kinematics. Example: 2-Link Structure. Forward Kinematics. Forward Kinematics

Capacity Utilization Metrics Revisited: Delay Weighting vs Demand Weighting. Mark Hansen Chieh-Yu Hsiao University of California, Berkeley 01/29/04

Geometrical Description of Signals GEOMETRICAL DESCRIPTION OF SIGNALS. Geometrical/Vectorial Representation. Coder. { } S i SOURCE CODER RECEIVER

Paul M. Sommers David U. Cha And Daniel P. Glatt. March 2010 MIDDLEBURY COLLEGE ECONOMICS DISCUSSION PAPER NO

CS 410/584, Algorithm Design & Analysis, Lecture Notes 5

Morningstar Investor Return

Lifecycle Funds. T. Rowe Price Target Retirement Fund. Lifecycle Asset Allocation

Time-Varying Correlations and Optimal Allocation in Emerging Market Equities for Australian Investors: A Study Using East European Depositary Receipts

An Alternative Mathematical Model for Oxygen Transfer Evaluation in Clean Water

The influence of settlement on flood prevention capability of the floodcontrol wall along the Bund in Shanghai

Overview. Do white-tailed tailed and mule deer compete? Ecological Definitions (Birch 1957): Mule and white-tailed tailed deer potentially compete.

CHAPTER TEST REVIEW, LESSONS 4-1 TO 4-5

Reduced drift, high accuracy stable carbon isotope ratio measurements using a reference gas with the Picarro 13 CO 2 G2101-i gas analyzer

Do Competitive Advantages Lead to Higher Future Rates of Return?

Research Division Federal Reserve Bank of St. Louis Working Paper Series

Risk analysis of natural gas pipeline

AP Physics 1 Per. Unit 2 Homework. s av

High Speed 128-bit BCD Adder Architecture Using CLA

Name Class Date. Step 2: Rearrange the acceleration equation to solve for final speed. a v final v initial v. final v initial v.

Urban public transport optimization by bus ways: a neural network-based methodology

Productivity and Competitiveness: The Case of Football Teams Playing in the UEFA Champions League

2. Literature Review Theory of Investment behavior

Bill Turnblad, Community Development Director City of Stillwater Leif Garnass, PE, PTOE, Senior Associate Joe DeVore, Traffic Engineer

IBIS: ATestbed for the Evolution of Intelligent Broadband Networks toward TINA

Automatic air-main charging and pressure control system for compressed air supplies

San Francisco State University ECON 560 Fall Midterm Exam 2. Tuesday, October hour, 15 minutes

Using Rates of Change to Create a Graphical Model. LEARN ABOUT the Math. Create a speed versus time graph for Steve s walk to work.

Automated External Defibrillators DESIGNED FOR UNEXPECTED HEROES

Betting Against Beta

The Comparison of Outlier Detection in Multiple Linear Regression

Authors: Christian Panzer, Gustav Resch, Reinhard Haas, Patrick Schumacher. Energy Economics Group, Vienna University of Technology

Identification of trend patterns related to the dynamics of competitive intelligence budgets (the case of Romanian software industry)

Brand Selection and its Matrix Structure -Expansion to the Second Order Lag-

A Critical Analysis of the Technical Assumptions of the Standard Micro Portfolio Approach to Sovereign Debt Management

Monte Carlo simulation modelling of aircraft dispatch with known faults

CALCULATORS: Casio: ClassPad 300 ClassPad 300 Plus ClassPad Manager TI: TI-89, TI-89 Titanium Voyage 200. The Casio ClassPad 300

Performance Optimization of Markov Models in Simulating Computer Networks

A Probabilistic Approach to Worst Case Scenarios

Analyzing the Bullwhip Effect in a Supply Chain with ARMA(1,1) Demand Using MMSE Forecasting

BISI Wear Dance Art Clothing. September 10, April 27, 2019

The t-test. What We Will Cover in This Section. A Research Situation

TRACK PROCEDURES 2016 RACE DAY

Interpreting Sinusoidal Functions

SELF-ASSESSMENT - MODULE 3-5: Gas Movement

Avoiding Component Failure in Industrial Refrigeration Systems

2017 MCM/ICM Merging Area Designing Model for A Highway Toll Plaza Summary Sheet

Wladimir Andreff, Madeleine Andreff. To cite this version: HAL Id: halshs

M.H.Ahn, K.J.Lee Korea Advance Institute of Science and Technology 335 Gwahak-ro, Yuseong-gu, Daejeon , Republic of Korea

Reliability Design Technology for Power Semiconductor Modules

Guidance Statement on Calculation Methodology

EFFECTS OF WIND SPEED ON WIND TURBINE AVAILABILITY

Evaluation of a Center Pivot Variable Rate Irrigation System

Strategic Decision Making in Portfolio Management with Goal Programming Model

Semi-Fixed-Priority Scheduling: New Priority Assignment Policy for Practical Imprecise Computation

3.00 m. 8. At La Ronde, the free-fall ride called the Orbit" causes a 60.0 kg person to accelerate at a rate of 9.81 m/s 2 down.

RECOMMENDATION FOR INTERCHANGEABLE STUD BOLTS AND TAP END STUDS FOR API SPEC 6A FLANGES

ADDITIONAL INSTRUCTIONS FOR ISU SYNCHRONIZED SKATING TECHNICAL CONTROLLERS AND TECHNICAL SPECIALISTS

Making Sense of Genetics Problems

Dynamics of market correlations: Taxonomy and portfolio analysis

ITRS 2013 Silicon Platforms + Virtual Platforms = An explosion in SoC design by Gary Smith

Transportation Research Forum

What the Puck? an exploration of Two-Dimensional collisions

3.10 Convected Coordinates

Proportional Reasoning

LEWA intellidrive. The mechatronic All-in-One pump system. intelligent flexible dynamic high precision. Foto: ratiopharm

NoC Essentials. Networks on Chips: A New Paradigm Module. Origins of the NoC Concept. Evolution or Paradigm Shift?

PERSONAL VERSION. Readers are kindly asked to use the official publication in references.

Real-time Stochastic Evacuation Models for Decision Support in Actual Emergencies

System GMM estimation with a small sample Marcelo Soto July 15, 2009

KINEMATICS IN ONE DIMENSION

I t ' 4 ti. t ti. IQ:::: mass x heat of fusion (or heat of vaporization) I HEAT AND ITS MEASUREMENT. t t. t f I I I. Name

XSz 8... XSz 50 Solenoid actuated fail-safe safety valve

INSTRUCTIONS FOR USE. This file can only be used to produce a handout master:

VLSI Design I; A. Milenkovic 1

Corresponding Author

2. JOMON WARE ROPE STYLES

A PROBABILITY BASED APPROACH FOR THE ALLOCATION OF PLAYER DRAFT SELECTIONS IN AUSTRALIAN RULES

Evolutionary Sets of Safe Ship Trajectories: Evaluation of Individuals

3. The amount to which $1,000 will grow in 5 years at a 6 percent annual interest rate compounded annually is

SURFACE PAVEMENT CHARACTERISTICS AND ACCIDENT RATE

The Yen and The Competitiveness of Japanese Industries and Firms. March 4, 2008 (preliminary draft) Robert Dekle Department of Economics USC

EXAMINING THE FEASIBILITY OF PAIRED CLOSELY-SPACED PARALLEL APPROACHES

VLSI Design I; A. Milenkovic 1

Engineering Analysis of Implementing Pedestrian Scramble Crossing at Traffic Junctions in Singapore

The Impact of Demand Correlation on Bullwhip Effect in a Two-stage Supply Chain with Two Retailers

Development of Accident Modification Factors for Rural Frontage Road Segments in Texas

Economic Growth & Development: Part 3 Horizontal Innovation Models: Applications. By Kiminori Matsuyama. Updated on , 6:12:26 PM

A computational model to assess the impact of policy measures on traffic safety in Flanders: Theoretical concepts and application.

Transit Signal Priority: Help or Hype? Peter G. Furth Northeastern University

Explore Graphs of Linear Relations. 1. a) Use a method of your choice to determine how much water will be needed each day of a seven-day cruise.

Economics 487. Homework #4 Solution Key Portfolio Calculations and the Markowitz Algorithm

Homework 2. is unbiased if. Y is consistent if. c. in real life you typically get to sample many times.

Transcription:

NoC Impac on Desgn Mehodology Avnoam Kolodny NoC as Means o Handle Complexy (Oucome of Moore s Law) Prncples for dealng wh complexy: Absracon Herarchy egulary Desgn Mehodology 000 00 Termnals per module (T) 0 en s rule: T = k N r en s exponen < Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 0 0 0 0 2 0 3 0 4 Transsors per module (N) 0 5 0 6 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 2 Characerscs of NoC-based desgn Each sysem module can sll be desgned n he radonal way The dfferences are a he full-chp sysem level: Sysem negraon nvolves nework synhess A sysem-level raffc model s requred uno us a connecvy nels placemen: performed n he conex of nework opology uno us geomercal placemen udrven by power and delay consderaons Transacons a he sysem level ake mulple clock cycles Sharng of nework resources nroduces unceranes usascal mehods for NoC desgn are evolvng Many ssues ranslae no nework resource allocaon problems Le s look a an example: sysem mng closure Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 3 Tmng Closure n Tradonal VLSI Inerconnec-based mng analyss s n a feedback-loop Synhess mus predc he delays somehow Wreload models are used for pre-layou C esmaon The eraon loop mgh no converge feedback Logc Synhess Schemacs (nels) Sac Tmng analyss feedback Layou Synhess Layou C exrac Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 4

Tmng closure n NoC Placemen n a NoC Defne nermodule raffc Defne nermodule raffc Place modules Place modules Increase lnk capaces Allocae lnk capaces QoS Sasfed? In a NoC, mng convergence becomes a nework capacy allocaon problem Verfy QoS and cos Too low capacy resuls n poor QoS Too hgh capacy wases power/area Unform lnk capaces are a wase n applcaon-specfc sysems! Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 5 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 6 Ierave allocaon of lnk capacy Defne nermodule raffc Problem Essence How much capacy [bs/sec] should be assgned o each lnk? All daa flows mus mee her delay requremens Mnmze oal resources Place modules Smulaon for mng verfcaon akes oo long..and s daa-dependen Allocae lnk capaces Verfy QoS and cos A smple model o esmae delays hrough he nework s requred! Need Sac Tmng Analyss for he nework Need he delay for each of he daa flows Consder effecs of wormhole swchng Consder neracon among flows Consder vrual channels Do no assume all lnks are equal Too low capacy resuls n poor QoS Too hgh capacy wases power/area Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 7 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 8

Wormhole Delay Analyss Packe Delvery Tme The delvery resembles a ppelne pass Packe ransmsson can be dvded no wo separaed phases: Pah acquson Fls ransmsson IP2 The delvery me of long packes s domnaed by he slowes lnk Transmsson rae Lnk sharng IP2 Low-capacy lnk We focus on he fls ransmsson phase IP IP Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 9 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 0 Packe Delvery Tme Analyss Bascs The delvery me of long packes s domnaed by he slowes lnk Transmsson rae Lnk sharng IP2 IP3 Deermne he flow s delay n each lnk Accoun for nerleavng IP Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 2

Sngle Hop Flow, No Sharng Sngle Hop Flow, wh Sharng = C l = C Λ l Bandwdh used by oher flows on lnk - mean me o delver a fl of flow over lnk [sec] C - capacy of lnk [bs per sec] l - fl lengh [bs/fl] C l Λ - mean me o delver a fl of flow over lnk - capacy of lnk [bs per second] - fl lengh [bs/fl] - oal fl necon rae of all flows sharng lnk, excep flow [fls/sec] Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 3 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 4 The Convoy Effec Emprcally consder ner-lnk dependences: Wormhole backpressure Traffc ams down he road Toal Packe Transmsson Tme Weakes lnk domnaes ransmsson me Accoun for all subsequen hops Λ l = + C ds (, k) k k k k π k Lnk Load Basc delay weghed by dsance Packe sze [fls/packe] T m max( π ) nework Accoun for weakes lnk Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 5 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 6

Queung occurs a he source Queung delay: T Q = 2 ( λ ) 2 T nework Add Queung Delay nework So, We have a sac delay model Emprcal formula based on average necon raes Easy o evaluae (non-smulaon) Provdes a reasonable esmae for average delay hrough he nework Now, we can use he model for allocang suffcen bandwdhs o he lnks of he NoC * I. Waler, Z. Guz, I. Cdon,. Gnosar and A. Kolodny, Effcen Lnk Capacy and QoS Desgn for Wormhole Nework-on-Chp, DATE 2006. Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 7 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 8 Lnk Capacy Allocaon Problem Use he delay analyss o solve an opmzaon problem Gven: Sysem opology and roung Each flow s bandwdh (f ) and delay bound (T EQ ) Mnmze oal lnk capacy Ce e E Such ha: Capacy Allocaon Algorhm Greedy, erave algorhm For each src-ds par: uuse delay model o denfy he mos sensve lnk uincrease s capacy uepea unl delay requremens are me flow : T T EQ Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 9 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 20

Capacy Allocaon Example# Unform raffc wh dencal requremens Unform allocaon: 74.4Gb/sec Capacy allocaon algorhm: 69Gb/sec Toal capacy reduced by 7% Capacy Allocaon Example#2 A SoC-lke sysem wh heerogeneous raffc demands and delay requremens Unform allocaon: 4.8Gb/sec Capacy allocaon Algorhm: 28.7Gb/sec Toal capacy reduced by 30% Before opmzaon Afer opmzaon 00 0 02 03 0 2 3 20 2 22 23 30 3 32 33 Before opmzaon Afer opmzaon 00 0 02 03 0 2 3 20 2 22 23 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 2 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 22 Slack [%] Slack Elmnaon Packe Delay Slack Exenson I: Analycal (non-emprcal) delay models Goal: Analycal delay model nsead of an emprcal one Approaches: Use nework calculus o compue delay bounds usee a hs conference: Analyss of Wors-case Delay Bounds for Bes-effor Communcaon n Wormhole Neworks on Chp, by Y. Qan e al. Analycal Modelng and Evaluaon of On-Chp Inerconnecs Usng Nework Calculus, by M. Bakhouya e al. Flow Slacks canno be pushed o zero because lnks are shared Use queung heory o oban sascal average delay per flow usee a hs conference: Packe-Level Sac Tmng Analyss for NoCs, by E. Krmer e al. Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 23 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 24

Queung heory approach: Transformaon no a fne Markov Chan Exenson II: Sascal Approach o NoC Desgn Gven: Traffc marx dsrbuon (consderng many dfferen use cases) Topology oung Lnk capaces Compue a sascal congeson guaranee 95% of raffc marces wll receve enough capacy * Packe-Level Sac Tmng Analyss for NoCs, by E. Krmer e al., Sesson 3 (poser ) Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 25 * I. Cohen, O. oensrech, and I. Keslassy, Sascal Approach o NoC Desgn,", NoCS '08 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 26 Traffc Load Dsrbuon Plo (T-Plo) Cavea: The Capacy Allocaon Paradox Gaussan? A Node A Node B C A C B ouer C Node C PDF 99.99% of raffc marces brng load under.6 Wors-case raffc load = 2 20% gan B Unlmed queues Fne (small) buffers Capacy Allocaon Paradox: Addng Capacy Can Desablze he Nework (reasons: small buffers + arbraon polcy) 27 Lnk Load * A. Baron,. Gnosar and I. Keslassy, "The Capacy Allocaon Paradox," IEEE Infocom '09 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 27 Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 28

Capacly Allocaon Paradox: Inuon Assume A has prory: (a) C A = (b) C A =2 [pk/t] [pk/t] Share of C 2 Share of C 2 Node A C A =2 A = Node B C B = A A B B () ouer Mn. Sze Buffer C =2 Node C Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 29 T A2 A2 B2 (a) 2T A3 B3 T/2 T 3T/2 2T 5T/2 (b) B (2) A3 3T B2 () 3T C B [Kf/s] C B [Kf/s] 500 450 400 350 300 250 200 50 00 50 Sably Pcure 0 C = 273[Kf/s] EP (Consan) 2 50 00 50 200 250 300 350 400 450 500 C A [Kf/s] C A [Kf/s] Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 30 4 3 2 0 A = B = 00[Kf/s] Concluson NoC-based mehodology helps handlng of nerconnecon complexy n large sysems Absracon Herarchy egulary Modulary Shared resources (lnks, buffers) New problems are seemngly nroduced by he NoC: Nework desgn opons/ssues esource sharng adds unceranes and farness ssues. Bu mos of hese problems are hdden somewhere anyway NoC-based desgn mehodologes are evolvng Excng research opporunes! Avnoam Kolodny NOC Desgn Tuoral NOCS 2009 3