Capacity Planning

1
Capacity Planning

• IACT 918 July 2004
• Gene Awyzio
• SITACS University of Wollongong

2
The Importance of Planning

• Network capacity planning lies at the heart of
  all network management functions
• The design of networks to route data from one
  location to another with the least cost is of
  great concern to network designers

3
The Importance of Planning

• The objective of network designers is to minimise
  the cost of installing and maintaining the
  network
• These problems are complicated by the need to
  consider many factors such as
• Cable capacity
• End-to-end blocking probabilities
• Delay and reliability requirements

4
The Importance of Planning

• The process of designing a network requires that
  many variables are determined simultaneously
• In designing and upgrading a network, it is
  necessary to simultaneously select
• Node location
• Links connecting backbone nodes
• Links between end users

5
The Importance of Planning

• Mathematically we can say that
• The objective of designing and maintaining a
  network is to
• Minimise the cost of installation and maintenance
• Whilst meeting some given performance criteria

6
The Importance of Planningor more specifically

• Given
• Node locations
• Edge locations
• Traffic requirementsbetween node pairs
• Cost of transmission capacity
• Cost of node installation
• Minimise
• Total network costs

• Determine
• Network topology
• Edge capacities
• Routing policy
• Subject to
• Performance constraints
• Capacity constraints
• Reliability constraints

7
Working It Out

• These models generally end up being
  combinatorially explosive or NP complete
• NP Non-deterministic Polynomial-time
• To simplify the problem of solving these models
  several assumptions are used
• Packet (data) arrival is independent of other
  traffic (Poissonian arrival)
• Packet size is independent of other traffic
  (exponential packet size)

8
Working It Out

• Modern network based applications dont follow
  these assumptions
• Most applications have a real time component
• These types of applications tend to create a
  traffic stream that creates packets
• At fixed time intervals (deterministic)
• Have a fixed size (deterministic)

9
Working It Out

• This means that all the simplifying assumptions
  are no longer valid
• Also,
• potential savings can be achieved by installing
  excess capacity to meet future requirements

10
Developing the Network Capacity Plan

• We start with a knowledge of the present network
  and its traffic loads etc, and our prediction of
  the future demands of the organisation
• We will develop the network capacity plan in
  three steps,
• facilities,
• equipment and
• evaluation

11
1. Facilities

• A facility is a transmission path between two or
  more locations, not including terminating or
  signalling equipment. The addition of
  terminating equipment would produce either a
  channel, a central office line, or a trunk.
• (Terplan P 556)

12
1. Facilities

• A facility has a theoretical maximum capacity
  which is given by the bandwidth
• In practice it is found to be impossible to use
  all the bandwidth because of
• Overheads
• Downtime
• Utilisation ceiling (depends on burstiness)

13
Facilities

• Capacity of facilities
• The utilization level beyond which the facility
  is unable to perform work in a timely manner
  within service-level limits
• The facility has reached its capacity when when
  its utilization has attained a level at which
  targeted service levels cannot be met.

14
1. Facilities

• Utilization ceiling
• The theoretical capacity of the facilities
• Identified by the speed of the resource
• How many bits can be transmitted per second
• Bits per second on average
• Bits per second at peak loading times
• There must be no compromise on service-levels for
  the average load
• Some violations at peak load may be tolerated
• Depends on the patterns of usage and the culture
  of the organisation

15
1. Facilities

• Terplan shows a graph as follows
• We will run out of capacity when we reach the
  usable limit. Then we are in trouble on this
  facility

Reserve
16
1. Facilities

• Utilization
• arrival rate / service time
• This yields a ratio (sometimes expressed as a
  percentage)
• gt1 unstable, queue will just grow bigger
• lt1 stable

17
1. Facilities

• As utilization increases in a stable system (?
  lt1), if the system were purely deterministic,
  then time would increase linearly
• But as this assumption does not hold, there is a
  knee point beyond which delays increase

idealized
Time
There is a knee often at around 6070
Utilization ?
18
1. Facilities

• Congestion collapse

perfect
Maximum carrying capacity of subnet
desirable
Packets delivered
congested
Packets sent
19
1. Facilities

• Response time service time waiting time
• Service time transmission time poll
  transmit time select transmit time
  poll/select wait times
• Waiting time (service time x utilization) / 2
  x (1 utilization)
• If response time lt response time objective, then
  the utilization ceiling is acceptableotherwise,
  the utilization ceiling must be reduced, by
  installing larger capacity lines, or more lines.

20
1. Facilities

• For practical purposes, we can estimate the
  effective capacity as about 60-70 of the
  bandwidth
• As the traffic approaches this maximum, the
  performance drops. Response time and packet loss
  increase
• This effective capacity is the upper limit
  available to the network, and we can use our
  estimates of future traffic to predict when we
  will run out of capacity

21
2. Equipment

• Equipment is everything not included in the
  facility
• Equipment capacity can be computed in the same
  way as facility capacity, using overhead,
  downtime and utilisation ceiling.

22
3. Evaluation

• When major upgrades are considered, there is
  always a choice among several alternatives
• The different possibilities must be developed and
  evaluated
• Development requires thinking of the
  alternatives, discussion among those involved and
  may include brainstorming

23
3. Evaluation

• Evaluation requires computer modelling or
  simulation
• It can be done on a spreadsheet or a simulation
  package designed for the purpose (eg Arena, nm)
• The value of any of these exercises depends on
  the accuracy of the original information on
  present and future loads