SoC Embedded NS Embedded Systems 3 Jan 2014

download SoC Embedded NS Embedded Systems 3 Jan 2014

of 63

Transcript of SoC Embedded NS Embedded Systems 3 Jan 2014

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    1/63

    Embedded (HW) Syste

    Performance Evaluatio

    CommunicationVineet [email protected]

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    2/63

    Outline Introduction

    System-on-a-Chip (SoC) Who enables Chip design

    Electronic Design automation (EDA) High level, Logic, Layout synthesis

    FPGA based synthesis

    ES/SoC Communication Modeling Performance Evaluation techniques

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    3/63

    Generic SoC Implementa

    Computat PEs

    Communi

    Bus bas NoC arc

    ?

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    4/63

    Generic SoC No. of PEs

    PE are verified, validated CPU, DSP , IP , ASIC Set-top-box, MPEG-2 decoder

    PE are heterogeneous Optimized HW/SW

    Separable computation & commuDiverse communication

    Proce

    L

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    5/63

    System Modeling Models of Computation

    Continuous Time Models SIMULINK, VHDL-AMS

    Discrete Timed Models

    HDLs etc. Synchronous Models Untimed Models

    Data Flow Process Networks Synchronous Data Flow, Kahn Proc

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    6/63

    Chip Design: Who Ena Design Technology

    Integrated circuit (Chip) design

    Applications technologies Wireless/Telecommunication

    Computing

    Internet Consumer/medical electronics

    Enabling technologies

    FabricationS f

    De

    Fabr

    k

    I

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    7/63

    Design Technology Systems characteristics Billion/Trillion transistors

    Heterogeneous, portable, low power, low Desired

    Circa 2004, A single hand-held portable syst

    Computing, GPS enabled, multimedia/voice/wireless internet

    Circa 2020, Similar system which is hands frwearable and controlled by thoughts

    VLSI DesignS h i b d h d l

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    8/63

    VLSI Design

    Methodology Supported by CAD

    Human expertisecentric

    Increasing mismatch of

    Synthesis models andimplemented circuit

    First time correctdesigns are rare/nearly

    Architectu

    Components

    System s

    Id

    Chip Lay

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    9/63

    Electronic Design Automation

    Architecture Synthesis

    Components/gate synthesis

    System specification

    Chip Layout synthesis

    SynthesisCAD tools

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    10/63

    Electronic Design Autom What ?

    Automating the design process

    Design automation using E-CA

    How ?

    Design description Abstraction, models

    Verification/simulation

    Transformation between various

    Back-annotatio

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    11/63

    E-CAD Why ?

    CAD is indispensable

    Design size too large

    Systems too complex

    Huge design efforts

    Problem definition dynamically c

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    12/63

    Performance Evalua

    Models for System-onCommunication

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    13/63

    Outline

    Communication Architectures (C

    Performance estimation Statistical estimation Analytical estimation

    Proposal for performance evaluamodels

    HCFG based approach

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    14/63

    Communication archite CA is fabric

    data & con It consist o

    1.Logic com

    2.Global bu3.Bus inter

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    15/63

    Bus based Single sha

    Hierarchicconnected

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    16/63

    Networks on Chip Packe

    via chswitch

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    17/63

    NoC TopologiesSPIN

    Scalable PrIntegrated N

    CLICH. Chip Level

    communica

    Heterogene

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    18/63

    Commercial Bus Archite IBM Core-Connect

    ARM AMBA Palmchip

    CoreFrame

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    19/63

    Performance Estimatio

    Statistical analysis

    - CAG model

    - Queueing theory

    - Stochastic framework Trace driven simulation

    - System level tools (PTO

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    20/63

    Efficiency of Modeling Appr

    Models Cl

    FSM, Timed-Petri-nets, Markov chains,

    Kahn process networks ,Transaction Levelmodels, OSM

    Si

    mo

    Synchronization graph, CAG, Queuingmodel

    Anmo

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    21/63

    Significance of interface synth

    Abstract Communication

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    22/63

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    23/63

    Alternate Comm. Architecture -

    C1,C2 & Mmapped to

    C3,C4 & Mmapped to

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    24/63

    Alternate mapping ( Arch-B) - C

    C1,C3 & Mmapped to

    C2,C4 & Mmapped to

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    25/63

    Queueing model

    Kim

    CAD

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    26/63

    Markov Model

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    27/63

    Hierarchical bus mode Indivi

    Ma Bus s

    Pas

    Pas

    Bridg

    Pas

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    28/63

    Hierarchical bus mode

    Pi1S1

    S

    2

    SG Introduction

    HCFG: An Overview

    HCFG h f l ti GSMP d l f SSB hit t

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    29/63

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    Evaluating Generalized Semi Markov Process

    Model of SoC Bus Architectures using HCFG

    Ulhas Deshmukh1 Vineet Sahula2

    1Govt. Polytechnic College, Dhule, India

    2Department of ECE

    Malaviya National Institute of Technology, Jaipur, India

    TENCON 2009, Singapore 23-26 November 2009

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    30/63

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    Outline

    1 Introduction

    Motivation

    Previous work and Proposed work

    2 HCFG: An Overview

    3 HCFG approach for evaluating GSMP model of SSB architecture

    SSB architectureGSMP Model formulation

    Model evaluation using HCFG approach

    HBB architectureGSMP model formulation

    Model evaluation using HCFG approach

    4 Results

    5 Conclusions

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architectureMotivation

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    31/63

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    Previous work and Proposed work

    Outline

    1 Introduction

    Motivation

    Previous work and Proposed work

    2 HCFG: An Overview

    3 HCFG approach for evaluating GSMP model of SSB architecture

    SSB architectureGSMP Model formulation

    Model evaluation using HCFG approach

    HBB architectureGSMP model formulationModel evaluation using HCFG approach

    4 Results

    5 Conclusions

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architectureMotivation

    http://find/http://goback/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    32/63

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    Previous work and Proposed work

    System level SoC architecture

    SW IP1

    On Chip Communication

    SW IP2

    HW IP1 HW IP2

    HW Adaption

    HW Adaption

    HW AdaptionHW Adaption

    CPU

    CPU

    SW Adaption

    SW Adaption

    SW

    SW SoC performs computation

    and communication of

    system functionalitySystem computation-

    Mapped to PEs/IPs

    Pre-verified & optimized

    Communication

    architecture facilitates

    communication among IPs

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architectureMotivation

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    33/63

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    Previous work and Proposed work

    Issue

    Modern-day systems demand for higher performance and

    more functions

    Designer has to use a number of concurrent and

    heterogeneous IPs

    Communication among these IPs becomes intricate

    Quick performance evaluation of communication

    architecture is essential

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architectureMotivation

    P i k d P d k

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    34/63

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    Previous work and Proposed work

    Outline

    1 Introduction

    Motivation

    Previous work and Proposed work

    2 HCFG: An Overview

    3 HCFG approach for evaluating GSMP model of SSB architecture

    SSB architectureGSMP Model formulation

    Model evaluation using HCFG approach

    HBB architectureGSMP model formulationModel evaluation using HCFG approach

    4 Results

    5 Conclusions

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architectureMotivation

    P i k d P d k

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    35/63

    pp g

    Results

    Conclusions

    Previous work and Proposed work

    Related work & proposed approach

    Classification Authors Approach

    Analytical P.Knudsen et al.[1] Delay estimation model

    S.Dey et al. [2] Synchronization graphT.Mudge et al. [3] SMP model (multi-bus arch.)A.Ramani et al. [4] SMP model (crossbar arch.)Proposed work GSMP model evaluation

    using HCFG

    Simulation J.Rowson et al. [5] Simulator (Cheetah)X.Zhu et al. [6] Operating State Machine

    Hybrid K.Lahiri et al.[7] Simulation + CAGS.Kim et al.[8] Queueing analysis + simulation

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    http://goforward/http://find/http://goback/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    36/63

    pp g

    Results

    Conclusions

    Hierarchical Concurrent Flow Graph (HCFG)

    HCFG approach [9] [10]

    An efficient analysis technique

    Captures concurrency, hierarchyHandles stochastic variation in task execution times

    Quickly compute average values & distribution

    Has been used for evaluation and improvement of process

    completion time of VLSI design processes [9] [10]

    [9] V.Sahula,C.P.Ravikumar,D. Nagchoudhuri,ASP-DAC 2002

    [10] V.Sahula and C. P. Ravikumar, VLSI Design,2001

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    http://-/?-http://-/?-http://-/?-http://-/?-http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    37/63

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    38/63

    Results

    Conclusions

    Communication concurrency in HCFG

    T T T

    TTT

    T T

    Ttile, SSource, Ddestination

    ST

    D

    P2 P3 TP1

    T2P 3

    TP

    P1

    Source PE

    Dest. PE

    Path 3Path 1 Path 2

    OR concurrency

    Multiple communications path can be active

    E[TC] =E[Min{TP1, TP2,TP3}]

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architectureSSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    39/63

    Results

    Conclusions

    HBB architecture

    Outline

    1 Introduction

    Motivation

    Previous work and Proposed work

    2 HCFG: An Overview

    3 HCFG approach for evaluating GSMP model of SSB architecture

    SSB architectureGSMP Model formulation

    Model evaluation using HCFG approach

    HBB architectureGSMP model formulationModel evaluation using HCFG approach

    4 Results

    5 Conclusions

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    40/63

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    41/63

    Results

    Conclusions

    Model formulation

    Arbiter

    BUS

    MEM

    PE PEPE1 2 N

    I/FI/F

    I/F I/FI/F Computing action

    ThePE1 is computing

    No request is generated

    Modeled ascomputing state

    (labeled asstate 0)

    Mean sojourn time is0 =T

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    42/63

    Results

    Conclusions

    Model formulation cont.

    Arbiter

    BUS

    MEM

    PE PE PE1 N2

    I/FI/F

    I/F I/F I/F

    Accessing action

    PE1 generates a request

    Bus and memory are idle((1 BUSY)2)

    Request wins arbitration (WIN)

    PE1 accesses the memory

    Modeled asaccessing state(labeled asstate 1)

    Mean sojourn time is1 =C

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    43/63

    Results

    Conclusions

    Model formulation cont.

    MEM

    BUS

    CP

    P FWPE1

    PEk

    PCOMP, CCOMM

    FWFull Waiting

    PEPEPE1 2 N

    I/F

    I/FI/FI/F

    Full waiting action

    PE1 &PEkgenerate request

    Bus and memory are idle

    PE1 does not win arbitration

    PEkgets access to memory

    PE1 has to wait for full accessing

    time ofPEk

    Modeled asfull waiting state

    (labeled asstate 2)

    Mean sojourn time isC

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    44/63

    Results

    Conclusions

    Model formulation cont.

    MEM

    BUS

    PCOMP, CCOMM

    RWResidual Waiting

    PEk

    P PE1

    C

    RW

    PE PE PE N1 2

    I/F

    I/F I/F I/F

    Residual waiting state

    PEk is accessing memory

    i.e. memory is busy

    PE1 generates a requestPE1 force to wait for

    residual accessing time of

    PEk

    Modeled asresidual

    waiting state(labeled as

    state 3)

    Mean sojourn time is

    (C2 C)/(2(C 1)

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    45/63

    Results

    Conclusions

    Model input parameters forith PE

    Parameters Description

    N number of processing elements

    Ti mean computing time

    Ci mean communication time

    C2i second moment of communication time

    Terms used in model formulation

    BUSY probability that memory or bus busyWIN probability of wining arbitration

    k mean sojourn time ofkth state

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://goforward/http://find/http://goback/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    46/63

    Conclusions

    GSMP model of a PE in SSB

    3

    0

    2 1

    3

    1

    1

    0

    1

    2

    11

    States of a PE

    State 0-computing stateState 1-accessing state

    State 2-full waiting state

    State 3-residual waiting

    state

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    47/63

    Conclusions

    HCFG based evaluation approach

    Transformation of GSMP model to HCFG

    Add extra nodeSfor initial task

    Draw directed edge(S, x),xis state of GSMP

    xshould have zero in-degree (minimum),pSx=1

    Mapjs of GSMP model to the Tjs of HCFG node,j=0,1, 2,3,Ts=0

    Mappij

    of GSMP model, to the weightpij

    of an edge (i,j) ofHCFG

    Consider one state at a time as a final (F) state

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    48/63

    Conclusions

    HCFG based evaluation approach cont.

    HCFG of a PE in SSB for N=2

    2p Z21

    1p Z10

    1p Z03

    1p Zs3

    p = 0.3921

    p = 0.6122

    p = 0.1802

    p = 1.0s3

    p = 0.6132

    p = 0.3931

    10p = 1.0 p = 0.43

    03

    p = 0.3901

    T = 0S

    T = 22

    T = 21

    T = 10

    T = 13

    232

    p Z

    2p Z02

    2p Z01

    2p Z31

    2p Z22

    3S 2F1

    0

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    49/63

    Conclusions

    Computation of steady state probabilities

    Describe HCFG in DFLOW tool

    Execute DFLOW description to find steady state probability

    of being in state F

    Repeat for all other states and for varing number of PEs

    Performance parameters are computed as follows:

    BW = N P1

    PU = P0+ P1

    L = N(P2+ P3)

    W = 22 + 33

    1

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    50/63

    Conclusions

    Outline

    1 Introduction

    Motivation

    Previous work and Proposed work

    2 HCFG: An Overview

    3 HCFG approach for evaluating GSMP model of SSB architectureSSB architecture

    GSMP Model formulation

    Model evaluation using HCFG approach

    HBB architecture

    GSMP model formulationModel evaluation using HCFG approach

    4 Results

    5 Conclusions

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    C l i

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    51/63

    Conclusions

    HBB architecture

    BusI/f

    BusI/f

    Bridge

    Arbiter ArbiterMEM1 MEM 2

    BUS1 BUS2

    I/F I/F

    I/F

    PE2

    I/F

    PEN

    I/F

    PE1

    I/F

    PE2

    I/F

    PEN

    I/F

    PE1

    I/F I/F

    Concurrent communication onBUS1&BUS2

    Each PE can accessMEM1&MEM2

    ConsiderPE1 mapped toBUS1

    MEM1-local memory,MEM2-global memory

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Concl sions

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    52/63

    Conclusions

    GSMP Model of a PE in HBB

    2 1

    46 5

    0

    1

    4

    0

    0

    2

    1

    4

    4

    4

    11

    6

    4

    5

    13

    3

    States of a PE

    State 0 computing state

    State 1 through state 3localMEM1

    State 4 through state 6

    globalMEM2

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    53/63

    Conclusions

    Model input parameters

    N = number of Processing Elements (PEs)

    T= Mean value of of think time

    X

    = Probability of local requestXg= Probability of global request

    C = Mean value of of local connection time

    Cg= Mean value of of global connection time

    C2l/C2g= Second moment of local/global communicationtime

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    IntroductionHCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    54/63

    Conclusions

    Model evaluation- HCFG approach

    HCFG of a PE in HBB forX=0.1.

    21p = 0.27

    10p = 1.0

    22p = 0.73

    31p = 0.27

    65p = 0.82

    40p = 1.0

    55p = 0.82

    54

    p = 0.1864

    p = 0.18

    2p Z01

    06

    p = 0.19

    04p = 0.16

    32p = 0.73

    05p = 0.57

    03p = 0.01

    01p = 0.027

    S0p = 1.0

    02p = 0.057

    2p Z312

    p Z22 2p Z

    04

    2p Z02

    1p Z

    10

    1p Z40

    2p Z65

    2p Z64

    2p Z05

    2p Z04

    2p Z

    54

    1p Z06

    1p Z03

    2p Z32

    2p Z55

    1p Zs0

    T = 10

    T = 22

    T = 13

    T = 24

    T = 25

    T = 0S

    T = 21

    T = 16

    2 1

    0

    456

    3

    S

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    Introduction

    HCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    SSB architecture

    HBB architecture

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    55/63

    Conclusions

    Performance metrics of HBB architecture

    Performance metrics

    BW = N P1BWg = N P4

    L = N(P2+ P3)

    Lg = N(P5+ P6)

    W = 2(2+31)/1+33Wg = 5(5+64)/4+66

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    56/63

    Introduction

    HCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    SSB architecture

    HBB architecture

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    57/63

    SystemC simulation setup for HBB architecture

    sc_indata_BR12;

    void bus2_action()

    SC_MODULE(bus2)

    {

    sc_indata_PE1;

    //PE11.cc

    while(true)

    sc_indata_BR12;

    void bus2_action()

    SC_MODULE(bus2)

    {

    sc_indata_PE1;

    //bus1.cc

    while(true)

    {

    }

    //arbiter1.cc

    SC_MODULE(arbiter2)

    { sc_inbus2_status;

    sc_inreq_PE1;

    //some ports

    void arbiter2_action() {

    while(true){

    //rest of the code not shown}

    }

    SC_CTOR(arbiter2)

    SC_THREAD(arbiter2_action);

    };

    {

    }

    //arbiter2.cc

    SC_MODULE(arbiter2)

    { sc_inbus2_status;

    sc_inreq_PE1;

    //some ports

    void arbiter2_action() {

    while(true){

    //rest of the code not shown

    }}

    SC_CTOR(arbiter2)

    SC_THREAD(arbiter2_action);

    };

    MEM 1 Arbiter1

    PE 1 PE 2

    MEM 1 Arbiter1

    PE 1 PE 2

    Bus

    I/f

    Bus

    I/f

    Bridge

    //some other ports

    if(PE_grant.read()==1)

    {

    //rest of the code not shown; }

    SC_CTOR(bus2)

    }

    };

    {

    {

    status_bus2.write(true);

    }

    }

    {

    //some other ports

    if(PE_grant.read()==1)

    {

    //rest of the code not shown; }

    SC_CTOR(bus2)

    }

    };

    {

    {

    status_bus2.write(true);

    }

    }

    {

    //BR1.cc

    SC_MODULE(BR1){

    //portssc_out req_BR1;

    sc_outdata_BR1;

    sc_in_clk clock;

    //other ports

    void BR1_action()

    while(true)

    { if(BR1_grant.read()==1)

    //rest of the code;}

    SC_CTOR(BR1)

    {

    SC_THREAD(BR1_action);

    sensitive

  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    58/63

    Results- SSB architecture

    Model Input Parameters -C=2 Cycles,T=2 Cycles,Cv=0.

    1 2 3 4 5 6 7 80

    0.2

    0.4

    0.6

    0.8

    1

    Number of PEs

    Bandwidth

    Simulation Approach

    AFOME Approach

    HCFG Approach

    1 2 3 4 5 6 7 80

    0.2

    0.4

    0.6

    0.8

    1

    Number of PEs (N)

    Pro

    cessorUtilization

    (PU)

    Simulation ApproachAFOME Approach

    HCFG Approach

    PU for T=2,r=0.1

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    Introduction

    HCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    59/63

    Results- HBB architecture

    Model Input Parameters - N=2,T=2 Cycles,C=Cg=2 Cycles,

    Cv=0.

    0.2 0.4 0.6 0.80

    0.2

    0.4

    0.6

    0.8

    1

    Probability of local request

    Localbandwidth

    Simulation Approach

    AFOME ApproachHCFG Approach

    0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.90

    0.5

    1.0

    1.5

    2.0

    2.5

    3.0

    3.5

    4.0

    Probability of local request

    L

    ocalwaitingtime

    Simulation Approach

    AFOME Approach

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    Introduction

    HCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    60/63

    Results- HBB architecture

    Model Input Parameters - N=2,T=2 Cycles,C=Cg=2 Cycles,

    Cv=0.

    0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.90

    0.1

    0.2

    0.3

    0.4

    Probability of local request

    G

    lobalbandwidth

    Simulation Approach

    AFOME Approach

    HCFG Approach

    0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.90

    0.5

    1.0

    1.5

    2.0

    Probability of local request

    G

    loalqueuelength

    Simulation Appro.AFOME Approach

    HCFG Approach

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    Introduction

    HCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    61/63

    Conclusions

    Proposed analytical approach is time efficient, accurateBesides, it gives stochastic properties quickly as that of

    simulation

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    Introduction

    HCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    62/63

    Bibliography

    1 P. Knudsen and J. Madsen, Integrating communication protocol selection with partitioning inhardware/software codesign in Proc. Int.Symp. Syst. Level Synthesis,1999,111-116.

    2 S. Dey and S. Bommu, Performance analysis of a system of communicating processes, in ICCAD, 1997,590-597.

    3 T.N.Mudge and H.B.A.Sadoun, A Semi Markov Model for Performance of Multiple-Bus Systems, IEEETrans. On Computers, Vol C-34, No 10, Oct 1985.

    4 A.K.Ramani, P.K.Chande and P.C.Sharma, A general model for performance investigations of prioritybased multiprocessor system, IEEE Trans. on Computers, 1992, 747-754

    5 J. A. Rowson and A. Sangiovanni-Vincentelli, Interface based design, Proc. Design Automation Conf.,1997, 178-183.

    6 X. Zhu, W. Qin, and S. Malik, Modeling operation and microarchitecture concurrency for communicationarchitectures with application to retargetable simulation, IEEE Trans. VLSI Syst., 14(7):707-716, July 2006.

    7 K.Lahiri, A. Raghunathan and S. Dey, System-level performance analysis for designing on-chipcommunication architecture, IEEE Tran. CAD, 20(6):768-783, June 2001.

    8 S. Kim, C.Im and S. Ha, Schedule-Aware Performance Estimation of Communication Architecture forEfficient Design Space Exploration, IEEE Trans. VLSI System Vol 13, No 5, Pages 539-552, may 2005.

    9 V.Sahula, C. P. Ravikumar and D. Nagchoudhuri. Improvement of ASIC Design Processes. ASP-DAC,2002, pages 105-112.

    10 V.Sahula and C. P. Ravikumar, The hierarchical concurrent flow graph approach for modeling and analysisof design processes, in Int. Conf. VLSI Design,2001,91-96.

    11 U.Deshmukh and V. Sahula, Analytical performance estimation from GSMP model for hierarchical busbridge based SoC communication architecture, in Proceedings of 20th IEEE International Conference onMicroelectronics (ICM), Dec. 2008.

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    Introduction

    HCFG: An Overview

    HCFG approach for evaluating GSMP model of SSB architecture

    Results

    Conclusions

    http://find/
  • 8/10/2019 SoC Embedded NS Embedded Systems 3 Jan 2014

    63/63

    Thanks...

    Ulhas Deshmukh, Vineet Sahula GSMP model evaluation using HCFG

    http://find/