Documente Academic
Documente Profesional
Documente Cultură
Presentation Outline
Planning of an ATM network Concepts of
Traffic streams Chains and routes Equivalent bandwidth Virtual paths and virtual circuits
Semester 2 - 2005
Slide 2
Learning Objectives
You will be able to:
Perform a simple computation of equivalent bandwidth Apply the Kaufman-Roberts formula Show that the KR formula is a simple extension of the Erlang model
Semester 2 - 2005
Slide 3
Introduction
ATM networks need to carry a wide variety of different traffic types broadband and narrow band services. Due to the non-homogeneity of this traffic and the difficulties in adequately characterising this traffic, the dimensioning of ATM networks is quite a challenge. Over-provisioning is a conservative but costly solution for ATM network planners. A number of models have been proposed for solving this problem. We shall present an overview of an approach involving the Kaufman-Roberts formula.
Semester 2 - 2005
Slide 4
Traffic Streams
Each traffic stream is defined as a single class of traffic between a pair of nodes (an OD pair). We model the stream arrival processes at the call level using Poisson processes. Each stream i is defined by its call arrival rate (i), call service rate (i), peak rate (Rpeaki), mean rate (mi) and average burst length (bi). The last three parameters given above define the onoff bursty cell arrival process within a call.
Semester 2 - 2005
Slide 5
Chains / Routes
A chain or route is defined as a directed sequence of links in a network that traces a path from an origin node to a destination node. A call will be accepted if there are capacities left to establish a virtual circuit connection between the origin and the destination.
Semester 2 - 2005
Slide 6
VPCx
VC Switch
CPE
VPCy
VP Switch VC Switch
VPCz
CPE
VCa
Semester 2 - 2005
VCb
Advanced Telecommunications 143.466
VCd
Slide 7
Other information required for call set-up (incl. supplementary service req.)
The terminal equipment converts this information into a Q.2931 call set-up message which is transmitted to originating local VC switch.
Advanced Telecommunications 143.466 Slide 8
Semester 2 - 2005
Call Model
Once the call has been accepted, it will generate cells at a peak rate Rpeak during an active (on) period and no cells during an inactive (off) period. The times in the on and off periods are negative exponentially distributed. By modelling the traffic as an on-off process, the method of Guerin et al is then used to determine the equivalent bandwidth of each traffic stream. Note that is is common to convert this equivalent bandwidth into slots that are multiples of 64kbps.
The slots can then be used in the Kaufmann-Roberts model.
Semester 2 - 2005 Advanced Telecommunications 143.466 Slide 9
1,1,Rpeak1,m1,b1 2,2,Rpeak2,m2,b2
N1
N2
N3
Semester 2 - 2005
Slide 10
Semester 2 - 2005
Slide 11
1 p ( m) = 0 1 K p( m ci )ci i i m i=1
:m = 0 :m < 0 :0 < m M
The above probabilities are then normalised by the following simple process:
Advanced Telecommunications 143.466 Slide 12
Semester 2 - 2005
p( j )
j =0
The blocking probability for the i-th class of call can then be computed as:
Bi =
m = M ci +1
p (m)
Semester 2 - 2005
Slide 13
For each link, find all streams that utilise it Using inverse Kaufman-Roberts, find the bandwidth for the call GOS
Slide 14
Dimensioning overview
( i , i , Rpeak , m , b )
Input Parameters call GOS buffer size cell loss probability chain information traffic parameter
Output bandwidth needed for each link in Mbps and number of 64Kbps slots
Slide 15
Equivalent Bandwidth
Slide 16
Equivalent Bandwidth
Cf =
Equivalent Bandwidth
Cs = m + '
where
N i =1
mi
N i =1
i2
Slide 18
where
i 2 = mi ( R peak i mi )
Inverse Kaufman-Roberts
Inverse Kaufman-Roberts is used to find the number of circuits when the GOS of multi-service streams is specified.
Mean =
i
i ci i
i , i , ci
loss = mean*GOS n?
Slide 19
Link Dimensioning
For each link, find all the traffic streams which utilise it. Taking the network GOS as the link GOS, and the traffic streams as input, use inverse KaufmanRoberts to find the number of circuits needed.
Slide 20
Slide 21
Slide 22
Iteration
Step 2
Calculate the GOS (g1) for lower bound (N1) and GOS (g2) for upper bound (N2) using Kaufman-Roberts formula The new N is set to be
ln (GOS/g2) ln (g2/g1)
Slide 23
Iteration
Step 2 (cont) It is found that the exponential fitting of curve is much more efficient than the linear interpolation method.
Slide 24
Iteration
Step 3 Using the Kaufman-Roberts formula, find the corresponding GOS (g) for N.
Iteration
Step 3 (cont) Repeat until upper bound = lower bound + 1
Then set
final circuit = upper bound
Slide 26
Conclusions
We have described a method for performing network dimensioning using the Kaufman-Roberts approach. This is not necessarily the best model for performing this dimensioning activity. We have not described how to perform optimal dimensioning
Semester 2 - 2005
Slide 27