Topological Optimization of Computer Networks

1
Topological Optimization of Computer Networks

• 990278 Yang,Jung Sook

2
Content

• Approach
• Overview
• Analyses
• book introduction
• Performance criteria
• Basic design techniques for centralized networks
• Distributed network design problems
• Conclusion

3
Approach

• Read
• Topological Optimization of Computer Networks
• By Howard Frank and Wushow Chou
• The world of large scale systems edited by James
  D. Palmer Richard Saeks

4
overview

• Main issue
• Modeling
• Analysis
• Design for
• centralized
• distributed
• computer-communication networks
• Basic problem To specify the location of each
  communication link within the network
• Design objective to provide
• Low-cost network
• Satisfy constraints
• Response time
• Throughput
• Reliability
• And other parameters..

5
Book introduction

• NASDAQ
• National Association of Securities Dealers
  Automated Quotation System
• System of computers and communication and
  terminal devices, designed primarily for the
  collection and distribution of real time
  quotations for the over-the-counter securities
  marker.
• NASDAQ central processing center
• Composed of low-, medium-, and high-speed
  communication lines
• Design constraints
• Small response time for several types of messages
• Extremely high reliability and availability
  specifications
• Low error rates on circuits
• The ability of the network to handle a large
  number of calls without loss of data

6

• ARPA
• Provides store-and-forward communications between
  a set of computers distributed across the US
• Message handling at center
• Through IMP(interface message processor)
• Connected by full duplex high-speed communication
  lines
• messages are broken into sets of packets
• Each packet independently makes its own way to
  its destination
• Goal of this system
• Average delay of less than 0.2s(per packet)
• Accommodate variations in traffic flow without
  degradation in performance
• Reliability at least two independent paths

7
NASDAQ and ARPA

• Different
• In philosophy of design and operation
• Similar
• The topological modeling and design
• to specify the capacity and location
• of each communication circuit
• within the network

8
Problems in topological optimization of network

• Network choices choosing M links
• Discrete Elements
• Discrete line speeds 2000,3400,240000 bits/s
• Integer optimization must be solved
• No theoretical method
• Nonlinearities
• Component cost structures, time-delay functions,
  reliability functions are nonlinear
• No analytical methods are available to obtain
  optimal solutions
• However, enormous progress
• this paper
• Describes a number of approaches to these
  problems
• indicates areas where additional research
• Suggests Possible approaches to several
  outstanding problem

9
Performance criteria

• The following questions must be answered
• Design constraints
• What average, maximum, or distribution of time
  delays will be tolerated?
• ARPA (0.2s)
• NASDAQ(less than 5s for over 50 of messages and
  less than 7s for over 90 of messages)
• What average and peak throughputs are required?
• ARPA 10 kbits/s on average , 85 on peak
• NASDAQ specifies busy hour traffic
• How should reliability be defined and what degree
  of reliability is necessafy?
• What types of communication modes (leased,
  point-to-point lines, direct-dial, multipoint
  polled, etc.)?

10
The goal of the network optimization

• The network design must provide solutions to the
  following problems
• the best message formats and transmission schemes
• kind of concentrator, multiplexer, and front-end
  configurations are needed, and location of them
• combination of leased lines, dial-up,
  point-to-point, multipoint, etc., and best speeds
  and locations
• the sensitivity of the designs performance to
  variations in traffic loads, equipment
  availabilities and other parameter changes

11
system models to solve the problems

• Time delay models and computations
• measure average delay time T by Queuing model
• Poisson arrival processes for all messages with
  an average r(j,k) messages per second and an
  exponential distribution of messages lengths with
  an average of 1/? bits per message. R is the sum
  of the quantities r(j,k), Ci is the capacity
  (bits/second) of the ith link, ??I is its average
  message traffic,B(the number of links), r(j,k)
• Pi propagation time on the ith channel. 1/?
  the average number of bits in an information
  message.

12
Basic design techniques for centralized networks

• Basic properties and heuristic structures
• -minimum cost solution
• optimal solution obtained by heuristic method.

13
Iterative techniques- centralized network

• Find lowest-cost tree
• Simple branch exchange procedure
• starts with any tree(each node has a unique path
  to the central node)
• Improvements by exchanging nodes on different
  paths(if the total cost is reduced, and exchange
  is feasible)

14
Distributed network design problems

• Iterative techniques for distributed network
  design
• Specify a feasible starting network
• Iteratively modifies the network in simple steps
  until a lower cost network satisfying all
  constraints is found.

15
Conclusion- system engineering view point

• Design philosophy - the structure of solution
• Definition of system engineering
• Transform an operational need into a description
  of system performance parameters and a system
  configuration through the use of an iterative
  process of definition, synthesis, analysis,
  design, test, and evaluation