Logical design: Network Management and Security

1
Logical design Network Management and Security

• Integrating Network Management and Security into
  the Design
• Defining Network Management
• Designing with Manageable Resources
• Network Management Architecture
• Security
• Security Mechanisms
• Security Examples
• Network Management and Security Plans

2
Example Security Breaches
3
Security points
4
The performance evaluation of computer networks

• Three different approaches
• Benchmarking
• Simulation
• Analytical Modelling

5
performance evaluation

• Benchmarking
• Build/modify the system
• measure the system while a standard set of tasks
  or workload is running.
• Merits
• Accuracy and direct applicability of the results
  obtained,
• ease of understanding and marketability of the
  approach

6
performance evaluation

• The demerits
• the high costs involved in obtaining the system,
  instrumentation,
• testing, large amounts of time and personnel
  involved,
• the results in many cases can not be extrapolated
  to suit changes in the system or the environment.

7
performance evaluation

• Simulation
• To build a simulation model of the system
• use of existing simulators (COMNET-III,
  NetCracker, OpNet, ns etc.) or write simulation
  programs in a general-purpose language such as
  C
• a limited amount of mathematical ability such as
  understanding random number generators etc. is
  needed.

8
performance evaluation

• Once the simulation program is ready and
  validated, one can run it with several alternate
  sets of parameter values to generate the
  performance measures in each case.
• Simulation models are flexible and have the
  property that arbitrary levels of detail can be
  generated.
• Excellent flexibility and range that can be
  simulated and detail that can be obtained are the
  merits.

9
performance evaluation

• The demerits
• long time and effort to construct simulators or
  simulation software.
• Though faster than benchmarking not as fast as
  analytical modelling. Usually a large number of
  long simulation runs are necessary for an
  elaborate and correct analysis.
• Sometimes that can be prohibitively large since
  the variable system parameters can be many.
• A new simulation run is usually necessary when
  there is a change in the system parameters or
  workload.

10
performance evaluation

• Analytical Modelling
• A mathematical model of the proposed system or
  part of it is constructed.
• This model is solved for exact or approximate
  solution using the cutting-edge mathematical
  techniques that are computationally efficient.
• Once this model is ready and validated, it can be
  used for performance evaluation and in several
  cases performance optimisation.

11
performance evaluation

• Perhaps this is the most promising approach, even
  if not for the present, definitely for the
  future.
• The advantages
• fast computations,
• possibility of optimisation and other analytical
  studies based on the formulae obtained.

12
performance evaluation

• The demerits
• the requirement of high level mathematical
  skills,
• several unrealistic assumptions involved,
• not as flexible or detailed as in simulation,
• not as accurate as in benchmarking.
• Analytical modelling is ideal for quick but
  approximate results. A lot of complex and useful
  analytical models are in the research stage and
  evolving.

13
Queuing Theory

• Queuing theory is used to estimate the
  performance of a number of facilities
• components of computer and communication systems
  such as
• computer processors,
• disks,
• terminals,
• data-access mechanisms,
• communication links,
• concentrators and
• memory can be modeled as queuing systems.
• (as well as manufacturing and transport systems)

14
Queuing theory

• A mathematical tool for describing, in
  mathematical terms, the behaviour of queues in a
  system
• Queuing theory is a mathematical tool for
  describing, in mathematical terms, the behaviour
  of queues in a system so that realistic estimates
  of response times and other values of interest
  can be computed.

15
Overview of Queuing Systems

• In computer and communication networks the
  contention of the resources results from the
  inability of a resource to service immediately
  all requests demanding service from it.
• The result is usually a build up of queues and a
  delay in obtaining service.
• In some cases (e.g. the telephone system)
  contention for resources may result in the
  rejection or 'blocking' of requests for service
  (i.e. the request leaves the system without
  receiving any service).
• In many design situations, there are three major
  constraints throughput, reliability and delay

16
Delay

• The time each Interface Message Processor (IMP -
  switching elements in networks) needs to store
  and forward a packet is the primary reason for
  networks when traffic is light
• Propagation delay for long distances
• As traffic increases queuing delay within each IMP

17
Queuing Theory

• Objective
• to design a system that is capable of providing
  the service demanded for example in terms of
  response time and maximum delay service-level
  objectives under varying levels of demand

18
Queuing Theory

• To achieve this, the behaviour of all the
  resources in the system must be understood in
  terms of likely service levels and delays under
  the range of predicted demand.
• Queuing theory is used to model as a queue any
  resource that is subject to contention.
• A queue is simply a service facility with a
  waiting room able to accommodate a line of
  requests (initiated by customers) waiting for
  service.

19
Queuing theory The Kendalls notation

• The Kendalls notation is used to describe the
  value of six parameters associated with a
  particular queuing system.
• A/B/c/K/m/Z

20
The Kendalls notation

• A The distribution of request arrivals. For
  communication networks it is the message packet
  arrival rate.
• B The distribution Of service times.
• For communication networks the server is simply
  the communication link and the service time is
  the link transmission time.
• For data networks B relates to the message or
  packet length distribution, and
• for voice networks the call duration distribution.

21
The Kendalls notation

• A and B refer to the characteristics of the input
  and output processes and are commonly described
  using the following symbols
• D Deterministic (constant) request arrival or
  service distributions (all customers have the
  same value)
• M Exponential (Markov) inter-arrival or service
  time distribution (i.e. Poisson arrival or
  service rate distribution).
• G General service time distribution (arbitrary
  probability density).

22
The Kendalls notation

• c The number of servers.
• For communication networks it is the number of
  communication links servicing the queue.
• K The maximum capacity of the queue.
• For communication networks this relates to the
  maximum queue length or buffer size.

23
The Kendalls notation

• m The Population of potential customers in the
  given source population.
• If the customer population is very large
  (practically infinite) the arrival of a customer
  at a queue will not affect the rate of subsequent
  arrivals. If the customer population is
  relatively small, (lt30), a single arrival will
  deplete the population and affect subsequent
  arrivals.
• Z The queuing discipline.
• Priorities, first come first served (FCFS) etc.

24
The Kendalls notation

• The queue discipline specifies the order in which
  arriving requests are placed in the queue for
  service (which implies the order in which they
  are serviced).
• Commonly used disciplines are FIFO (first-in,
  first-out) also known as FCFS (first-come,
  first-served), LIFO (last-in, first out), FIRO
  (first-in, random-out) or some form of priority
  discipline.

25
The Kendalls notation

• In Kendall's notation K, m and Z are commonly
  assumed to be equal to infinity and the notation
  shortened to
• A/B/c
• For example the simplest model, the M/M/1 model,
  describes a queue with
• a Poisson request arrival rate distribution,
• an exponential service time distribution and
• a single server.

26
The Poisson distribution

• The Poisson distribution is widely used in
  queuing theory since for many real-world systems
  (including communication systems)
• it approximates well to the actual behaviour of
  these systems.

27
The Poisson distribution

• In communication networks a message arriving is
  an event that is randomly occurring in time
• their duration is relatively short and
• the main characteristic of the event is the time
  of arrival.
• Message arrivals may also be described by
• the length of time between arrivals (the
  inter-arrival time) and
• the number of arrivals in a given time interval.
  
• A series of message arrivals in a time interval
  has no influence on arrivals in other
  non-overlapping time intervals.

28
The Poisson distribution

• Three basic assumptions are used to derive a
  message arrival rate based on the Poisson
  distribution
• Within a very short interval, the probability of
  only one message arriving in that interval is
  high.
• Arrivals are memoryless, i.e. arrivals of
  messages are independent of each other. This is
  likely to be the case when the messages are
  generated from a large number of independent
  sources.
• The characteristics of the message arrival
  distribution do not vary depending on the
  observation period.

29
The Poisson distribution

• In the case where in a service facility the next
  customer is served as soon as the one in service
  leaves the system, it is apparent that the time
  between service completions must be equal to the
  service time.
• Therefore if the time between completions is
  exponentially distributed, then the service time
  itself is exponentially distributed in time.
  Hence the service time distribution in this case
  is a Poisson process.

30
The Poisson distribution

• The Poisson process is a special case of a more
  general process known as the Markov process.
• A Markov process is a stochastic (random) process
  that exhibits a particular characteristic, namely
  that the distribution at any time in the future
  depends only on the current state of the process
  and not on how that state was reached (the
  memoryless property).

31
The Poisson distribution

• Other useful properties of the Poisson process
• a distribution resulting from the sum of Poisson
  distributions retains the Poisson distribution.
• if a Poisson stream is split into multiple
  substreams, each with a probability Pi of a job
  going to the ith substream, each substream is
  also Poisson with a mean rate of Pi.
• if the arrivals to a multiserver facility are
  Poisson with each server having exponential
  service times, the departures also constitute a
  Poisson stream.

32
Networks of queues

• A communication network consists of nodes that
  are interconnected by communication links.
• A communication network (in particular the
  backbone network) can thus be modelled as a
  network of queues.

33
The types of queuing network

• An open queuing network
• A closed queuing network

34
Open queuing network

• An open queuing network is one in which there are
  exchanges (arrivals and departures) of messages
  with the outside world.
• appropriate for modelling on-line transaction
  processing environments where the arrival rate
  does not depend on the response time perceived by
  the user.

35
Open queuing networks

• The use of these techniques has been shown to be
  capable of modelling device utilisations to
  within 10 and response times to within about 30
  of actual measured values. This is within
  acceptable limits given the typical accuracy of
  input data.

36
Closed queuing networks

• A closed queuing network involves no exchanges
  with the outside world
• A fixed number of messages internally circulate
  among the interconnected nodes.
• A closed queuing network is an open network with
  all source and destination rates set to zero.

37
Closed queuing networks

• For workloads such as batch and interactive
  systems in which a user continually interacts
  with the system over a long period of time,
  submitting a new request each time a reply is
  received, open queuing networks can not be used.
• In this case a closed queuing network model is
  appropriate.

38
Closed queuing networks

• In reality buffer memory is limited and leads to
  discarded messages or 'slowdown' where the
  arrival of messages into the system is restricted
  depending on the utilisation level of buffer
  memory.
• These and other situations are modelled as closed
  queuing networks and require more difficult
  mathematical analysis.

39
Queuing notation

• R facility utilisation l/m
• l average arrival rate
• m average service rate
• Es average service time
• N number of servers
• PD probability of delay
• Pk probability of k messages in queue

40
Queuing notation

• Em average number of messages waiting
• for service
• En average number of messages in the
• system
• ET average system delay time
• Ew average delay time

41
Network Simulation

• Develop a computer program to simulate a given
  system. The arrival rates follow a Markov
  Modulated Poisson Product distribution. Show all
  possible states of this system and associated
  transition states (i.e. arrival/departure rates)

42
(No Transcript)
43
Notation
44
Notation
45
States

• The state of the system at any point of time can
  be defined by a vector having the following
  values
• a,b,c,d,e,f

46
Possible states

• 1. An external arrival to server no 1 from
  outside changes the state to
• a1,b,c,d,e,f

47
Possible states

• 2. A departure from server no 1 to the outside of
  the system changes the state to
• a-1,b,c,d,e,f
• The departure takes place with a rate of