11.Computer Systems, Clusters, and Networks - Anvari.Net

clashjudiciousΗλεκτρονική - Συσκευές

8 Νοε 2013 (πριν από 3 χρόνια και 7 μήνες)

98 εμφανίσεις

Chapter 11

Modern Computer Systems, Clusters,
and Networks

The Architecture of Computer Hardware
and Systems Software:

An Information Technology Approach

3rd Edition, Irv Englander

John Wiley and Sons

㈰〳


Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
2

Basic Personal Computer System

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
3

Mainframe Computer System

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
4

Major PC System Components

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
5

System Performance
Improvements


Multiple CPUs


Faster clock speed, buses and circuits


Wider instruction and data paths


Faster disk access


More and faster memory

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
6

Multiprocessing


Reasons


Increase the processing power of a system


Parallel processing


Types of multiprocessor systems


Tightly coupled systems


Loosely coupled systems

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
7

Tightly Coupled Systems


Also called multiprocessor systems


Identical access to programs, data,
shared memory, I/O, etc.


Easily extends multi
-
tasking, and
redundant program execution


Two ways to configure


Master
-
slave multiprocessing


Symmetrical multiprocessing (SMP)

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
8

Tightly Coupled Systems

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
9

Master
-
Slave Multiprocessing


Master CPU


Manages the system


Controls all resources and scheduling


Assigns tasks to slave CPUs


Advantages


Simplicity


Protection of system and data


Disadvantages


Master CPU becomes a bottleneck


Reliability issues


if master CPU fails entire
system fails

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
10

Symmetrical Multiprocessing


Each CPU has equal access to resources


Each CPU determines what to run using a
standard algorithm


Disadvantages


Resource conflicts


memory, i/o, etc.


Complex implementation


Advantages


High reliability


Fault tolerant support is straightforward


Balanced workload

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
11

Loosely Coupled Systems


Clusters or multi
-
computer systems


Each system has its own CPU, memory, and
I/O facilities


Each system is known as a node of the
cluster


Advantages


Fault
-
tolerant, scalable, well balanced, distance is
not an issue


Two ways to configure


Shared
-
nothing model


Shared
-
disk model


Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
12

Shared
-
Nothing Model


High speed link between nodes


No sharing of resources


Partitioning of work through division of
data


Advantage


Reduced communication between nodes


Disadvantage


Can result in inefficient division of work

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
13

Shared
-
Disk Model


High speed link between nodes


Disk drives are shared between nodes


Advantage


Better load balancing


Disadvantage


Complex software required for
transactional processing (lock, commit
phases)


Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
14

Cluster Models

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
15

Beowulf Clusters


Simple and highly configurable


Low cost


Networked


Computers connected to one another by a private Ethernet
network


Connection to an external network is through a single
gateway computer


Configuration


COTS


Commodity
-
off
-
the
-
shelf components such as
inexpensive computers


Blade components


computers mounted on a motherboard
that are plugged into connectors on a rack


Either shared
-
disk or shared
-
nothing model

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
16

Blade and Rack of Beowulf Cluster

Figure 11.9

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
17

Computer Interconnection


Communication channel


pathway for data
movement between computers


Point
-
to
-
Point connectivity


Communication channel that passes data directly
between two computers


Serial connection


Telephone modem


Terminal controller


handles multiple point
-
to
-
point connections for a host computer


Multipoint connectivity


Multidrop channel or shared communication
channel


Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
18

Example: Point
-
to
-
Point

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
19

Client
-
Server Architecture


Computer servers provides services


File storage, databases, printing services,
login services, web services


Client computers


Execute programs in its own memory


Access files either locally or can request
files from a server

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
20

Client
-
Server Network

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
21

LAN Topology


Arrangement of workstations in a
shared medium environment


Logical arrangement (data flow)


Physical arrangement (cabling scheme)

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
22

LAN Topologies: Bus


Multipoint medium


Stations attach to linear medium (bus)
using tap


Transmission from any stations travels
entire medium (both directions)


Termination required at ends of bus to
prevent the signal from bouncing


Break in cable brings down entire bus

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
23

Bus LAN Diagram

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
24

LAN Topologies: Tree


Generalization of bus topology


Branching cable with no closed loops


Cable(s) begin at headend, travel to
branches which may have branches of
their own


Each transmission propagates through
network, can be received by any station

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
25

LAN Topologies: Ring


Repeaters are joined by unidirectional
point
-
to
-
point links in a ring


As data circulates past a receiver, the
receiver checks its address, and copies
those intended for it into a local buffer


Data circulates until it returns to source,
which removes it from network


Better performance at high levels of
usage


Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
26

Ring LAN Diagram

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
27

LAN Topologies: Star


Each station connected point
-
to
-
point to
a central station, usually with two
undirectional links


Switching in the central station connects
pairs of nodes together


Central node can broadcast info, or can
switch frames among stations


Failure of central station causes entire
network to go down

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
28

Star LAN Diagram

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
29

Ethernet MAC Protocol


MAC


Medium Access Control


Ethernet and CSMA/CD


Carrier sense multiple access with collision
detection


Four step procedure


If medium is idle, transmit


If medium is busy, listen until idle and then transmit


If collision is detected, cease transmitting


After a collision, wait a random amount of time
before retransmitting

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
30

Ethernet Frame

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
31

Switched Ethernet

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
32

Token Ring MAC Protocol


Token “seized” by changing a bit on the
circulating frame to indicate start of frame
rather than token


Default configuration requires sender to
complete transmission and begin receiving
transmitted frame before releasing the
token


“Early token release” allows release of
token after transmission but before receipt
of frame

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
33

Hubs


The active central element of the star
layout.


When a single station transmits, the hub
repeats the signal on the outgoing line
to each station.


Hubs can be cascaded in a hierarchical
configuration


Ethernet hubs are physically a star but
logically a bus.

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
34

Bridges


Allow connections between LANs and to
WANs


Used between similar networks


Read all frames from each network


Accept frames from sender on one network
that are addressed to a receiver on the other
network


Retransmit frames from sender using MAC
protocol for receiver

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
35

Routers


Similar to bridges but connect dissimilar
networks


Convert format of the message to
correspond to the protocol of the other
network


Network traffic is specifically addressed
to the router

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
36

Wide Area Network


Circuit switching


Dedicated channel between source and
destination for duration of connection


Message switching


Dedicated channel for an entire message


Packet switching


An independent path is created for each datagram


Virtual circuit switching


A route is created from source to destination
before transmission begins and all datagrams are
sent using the same route

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
37

Networks vs. Clusters


Externally, clusters appear as a single
computing unit.


Network nodes are individually
identifiable.


Workload on a cluster is determined by
cluster administration and load
-
balancing software.


Network workload cannot be controlled
using the above method.

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
38

High Performance Computing


Massively parallel processor architectures
(MPP)


Clusters of power machines or larger Beowulf
blade clusters


Well suited for problems that can be broken into
subtasks


Grid computing


Supercomputer performance through distributing
CPU processing to the spare CPU cycles of
personal computers connected to a network

Chapter 11: Modern Computer
Systems, Clusters, and Networks

11
-
39

Parallel Computers


Massively parallel architectures


Hundreds to millions of CPUs


CPUs have small amounts of local memory


All CPUs have access to global shared
memory


Pipelined CPUs


Results from one CPU flow to the next CPU for
additional processing