Scribd
Upload
31 views

Multiprocessors and Multicomputers

The document discusses different types of memory architectures in multiprocessor and multicomputer systems including shared memory, distributed memory, UMA, NUMA, COMA and CC-NUMA models. It also covers multiple generations of multicomputers and issues related to message passing architectures.

Uploaded by

Senthil Ganesh R
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
31 views

Multiprocessors and Multicomputers

The document discusses different types of memory architectures in multiprocessor and multicomputer systems including shared memory, distributed memory, UMA, NUMA, COMA and CC-NUMA models. It also covers multiple generations of multicomputers and issues related to message passing architectures.

Uploaded by

Senthil Ganesh R
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 27

MULTIPROCESSOR AND MULTICOMPUTERS

Type of memory access

• Shared Memory

• common shared memory

• Problem occurs when more than one process want


to write to (or read from) the same memory address

• Shared memory programming models do deal with


these situations
Distributed memory

• Networked processors with their private


memory
Shared Memory Multiprocessors

• UMA (uniform memory access)

• NUMA (non-uniform memory access)

• COMA (cache only memory architecture)

• Differ in how the memory and peripheral resources


are shared or distributed
The UMA Model
• Physical memory is uniformly shared by all the
processors

• All processors have equal access time to all


memory words – uniform memory access

• Processor may have private cache

• Peripherals also shared

• Multiprocessors- tightly coupled systems –high


degree of resource sharing
The UMA Model
• System interconnect takes the form of a common bus, a
crossbar switch

• Suitable for general purpose and time sharing


applications by multi users

• Used to speed up the execution of a single large


program in time –critical applications

• Synchronization and communication among processors


are done through using shared variables in common
memory
The UMA Model
• When all processors have equal access to all
peripheral devices- symmetric multiprocessor
• All processors are equally capable of running the
executive programs
The UMA Model
• Asymmetric multiprocessor – only one or subset of
processor are executive capable

• Master processor can execute the operating


system and handle I/O

• Remaining processors execute no I/O capability –


attached processors(APs)

• APs execute user codes under supervision of


master processor
The NUMA Model
• Shared memory access

• Access time varies with location of the memory word

• Shared memory is physically distributed to all


processors called local memories

• Collection of all local memories forms a global


address space accessible by all processors

• Faster to access local memory with local processor


The NUMA Model
• Access of remote memory attached to other processor
takes longer due to the added delay through the
interconnection network

• Globally shared memory –added to multiprocessor system

• 3 memory access patterns

• Fastest- local memory access

• Medium- global memory access

• Slowest- access of remote memory


The NUMA Model
• Processors are divided into several clusters

• Clusters are connected to global shared memory


module

• Entire system connected to NUMA

• Processor belonging to same cluster are allowed


to uniformly access the cluster shared memory
module

• All clusters have equal access to global memory


The NUMA Model
The NUMA Model
The NUMA Model
The COMA Model
• Multiprocessor using cache only model assumes
COMA Model

• Special model of NUMA , in which distributed main


memories are converted to caches

• No memory hierarchy at each processor node

• All cache form a global address space

• Remote cache access is assisted by the


distributed cache directions
The COMA Model
• Hierarchical directories –help locate copies of
cache blocks

• Data placement is not critical


CC-NUMA
• CC-NUMA –cache coherent non uniform memory
access
• Specified with distributed shared memory and
cache directories
Drawbacks of multiprocessor systems
• Suitable for general purpose multiuser applications

• Programmability is major concern

• Lack of scalability

• Difficult to build MPP machines using centralized


shared memory model

• Latency tolerance for remote access

• Packaging and cooling – constraint on scalability


Distribute memory multicomputers
• System has multicomputer computers(nodes)
interconnected by message passing network

• Each node – processor, local memory and disks or


i/o peripherals

• Message passing network provides point –point


static connection among the nodes

• Local memories are private – accessible only by


local processors
Distribute memory multicomputers

• Traditional machine –no remote memory


access(NORMA) Machines

• Internode communication – by passing message


through the static connection network
Distribute memory
multicomputers
Multicomputer Generations
• Use hardware routers to pass messages
• Computer node is attached to each router
• Boundary router is connected to i/o and peripheral
devices
• Message passing involves routers and channels
• Heterogenous multicomputer – mixed types of
nodes
• Internode communication –data representations
and
Multicomputer Generations
• First generation(1983-1987)
• Processor board technology
• Hybercube architecture
• Software controlled message switching

• Second generation(1988-1992)
• Mesh connected architecture
• Hardware message routing
• Software environment for medium grain
distributed computing
Multicomputer Generations
• Third generation multicomputer (1993-1997)
• Fine grain multicomputer
• Both processor and communication gears on
the same VLSI Chip
• Ring ,tree ,mesh ,torus ,hypercube ,cube
connected cycle –topologies in multicomputers
Issues in multicomputers
• Message –routing schemes

• Network flow control strategies

• Deadlock avoidance

• Virtual channels

• Message passing primitives

• Program decomposition techniques


Taxonomy of MIMD Computers
• Parallel computers are SIMD or MIMD
configuration

• SIMD-general purpose applications

• SIMD –not size scalable

• scalable multicomputers or multiprocessor-


distributed shared memory

• Unscalable multiprocessor-centrally shared


memory
Taxonomy of MIMD Computers
• Multicomputers use distributed memories with
multiple address space

• Scalable with distributed memory

You might also like