Tightly Coupled Microprocessors

download Tightly Coupled Microprocessors

of 14

Transcript of Tightly Coupled Microprocessors

  • 8/13/2019 Tightly Coupled Microprocessors

    1/14

  • 8/13/2019 Tightly Coupled Microprocessors

    2/14

    MIMD Systems

    Each PE Works

    independently ofothers.

    Processor 1 is said to beindependent of

    processors 2,3..N atany instant, if and only

    if task being executedby processor 1 has nointeractions with tasksexecuted by processor2,3.N and vice-versa.

    However results from

    processor X may beneeded by processorY.

    h

    r

    t

  • 8/13/2019 Tightly Coupled Microprocessors

    3/14

    This can be done by directlysending the results torequiring processors or

    storing them in a sharedmemory.

    Which causes the basis popular MIMD organizat

    (a) Shared Memory or TightArchitecture.

    (b) Message Passing or LoosCoupled Architecture.

  • 8/13/2019 Tightly Coupled Microprocessors

    4/14

    Shared Memory Architecture/Tightly CoupleArchitecture

    AnyProcessor

    i canaccess anymemory

    module jthroughinterconnec

    tionnetwork.

    The resultsof the

    computation are

    stored inthe

    memory bytheprocessor

    thatexecutedthe task.

    If thresultsrequireother ta

    they mbe accefrom memo

  • 8/13/2019 Tightly Coupled Microprocessors

    5/14

    Called TightlyCoupled

    Architecture, sincethe processors are

    interconnectedsuch that theinterchange ofdata between

    them through the

    shared memory isquite rapid.

    Memory accesstime is same forall processors

    hence also calledas Uniform

    MemoryArchitecture

    (UMA).

    If Processors aredifferent 32-bitand 16-bit and

    memory consists of32-bit words, eachmemory word mustbe converted intowords for the use

    of 16-bit

    processors. This isan Overhead.

    Another problem isMemory

    Contention whichoccurs when two

    or more processorstry to access same

    shared memoryblock.

    Since memoryblock can be

    accessed only bone processor a

    time, all theothers requestithe same blocmust wait untifirst processor

    through using i

  • 8/13/2019 Tightly Coupled Microprocessors

    6/14

    Message Passing Architecture/Loosely CoupleArchitecture

    There isno sharedmemoryat all in

    thesystem,

    each

    processorhas a localmemory

    blockattached

    to it.

    Theconglomeration

    of all localmemories is thetotal memory

    that the systempossesses. It is

    also calledDistributed

    MIMD.

    If daexchan

    requibetweeprocess

    thiconfiguthe requ

    processendmessage

    in whmemorreque

    data is s

  • 8/13/2019 Tightly Coupled Microprocessors

    7/14

    The requestedprocessor finishes itstask under process

    then access itsmemory for the

    requested data andpasses it on to

    interconnection

    network whichroutes it towards therequesting processor.

    For all this timerequesting processor

    sits idle incurringlarge overheads.

    Memory access timevaries between the

    processors andhence these

    architectures areknown as Non-

    Uniform Memory

    Access (NUMA).

  • 8/13/2019 Tightly Coupled Microprocessors

    8/14

    Comparison Of Tightly Coupled Vs. LooselyCoupled Architecture

    Tightly Coupled Architecture Loosely Coupled Architecture

    1. More rapid data interchange betweenprocessors in comparison to Loosely

    Coupled Architecture.

    1. Less rapid data interchange betweeprocessors in comparison to Tightly

    Coupled Architecture.

    2. More memory contention problem. 2. Less memory contention incomparison to tightly coupled

    architecture.

    3. Also called as shared memoryarchitectures, UMA, Multiprocessor

    system.

    3. Also called as message passingarchitectures, NUMA, Multicomputer

    system.

  • 8/13/2019 Tightly Coupled Microprocessors

    9/14

    Examples of Tightly Coupled Multiprocessors

    1. Cyber-170 Architecture

    Consists of

    twosubsystems

    (a) Central

    processingsubsystem

    (b)Peripheral

    processingsubsystem

    Both haveaccess to

    commoncentralmemory

    Through CMController,which is a

    high speedCross bar

    switch

  • 8/13/2019 Tightly Coupled Microprocessors

    10/14

    1. A Cyber-170 Multiprocessor Configurationwith two processors

    CM and ECM form two levelmemory hierarchy.

    CMC becomes switchingcentre and performs functionof ISIN, IOPIN and PMIN.

  • 8/13/2019 Tightly Coupled Microprocessors

    11/14

    2. Honeywell 60/66 Architecture

    Every centralprocessor and every

    I/O Mux is connectedto every systemcontroller (SC).

    This provides

    adequate redundancyin paths for highavailability.

    In event of failure on

    the SC, all IOMs arestill accessible byeach processor.

    SC acts a

    controassociamemor

  • 8/13/2019 Tightly Coupled Microprocessors

    12/14

    A Honeywell 60/66 multiprocessor system

    SC also acts as an intelligentswitch to route interruptsand other communicationsamong the various systemcomponents.

    When more than one elementattempts to access the samememory module, thecorresponding SC resolves theconflict.

    This triple redundancyenhances availability and

    fault tolerance.

  • 8/13/2019 Tightly Coupled Microprocessors

    13/14

    3. The PDP-10 multiprocessor

    There are two configurations ofPDP-10 multiprocessor systemwith multi-ported memorymodules. Each CPU has a cacheof 2K words, where each word is36 bits.

    Fig. shows Asymmetric Master-Slave configuration, the twoprocessors are identical, butasymmetry is result ofconnection of peripherals to themaster only.

    Slave can not initiate peripheraloperations nor respond to an

    interrupt directly.

  • 8/13/2019 Tightly Coupled Microprocessors

    14/14

    Fig. shows a symmetric configuration, bothprocessors are connected to a set of shared fastand slow peripherals, however each datachannel is attached to one processor, which isthe only processor that can use it.

    Slow peripherals are connected to both

    processors via a switch. There is no cache invalidate interface between

    them. It is assumed that a software solution isused to enforce cache consistency.

    Inherent redundancy in a multiprocessor systemprobably increases its ability to be faulttolerant.