Professional Documents
Culture Documents
on Loss Runlengths
Henning Sanneck and G. Carle
GMD Fokus, Berlin/Germany
fsanneck,carleg@fokus.gmd.de
MMCN 2000
January 25, San Jose, CA
Overview
Introduction
- Motivation
- Window-Based Mean Loss Rate
Intro
Motivation
Time-scale over which loss impact has to be assessed is still \short" however
Loss recovery mechanisms (FEC) are also sensitive to short-term loss
(short transmission groups ! short delay)
Need additional metrics to describe the loss process
Variety of models/metrics proposed in the literature
! Which metrics are useful ?
! Can a common framework for the models/metrics be identied ?
A Framework Model for Packet Loss Metrics Based on Loss Runlengths
Intro
p100(s)
p5(s)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
100
200
300
400
500
sequence number s
600
700
800
900
Long-term mean loss rate (m = 100) does not reveal dierences in perception
A Framework Model for Packet Loss Metrics Based on Loss Runlengths
Intro
analysis window
1
m
loss indicator
function
l(s)
1
s
1
pm(s)
Pa
l
( )wm (s
)
l
(s) wm(s)
=0
=
pm(s) =
m
m
s: sequence number, a: number of packets sent
A Framework Model for Packet Loss Metrics Based on Loss Runlengths
Intro
analysis window
1
s
loss indicator
function
l(s)
k=3
k=2
1
s
p (s)
1
00
p 20
X== 1
X=0
p 01
X== 2
p 12
p 23
X== m
pmm
p(m-1)m
m (model order) set e.g. to lowest number of packets for which an audio
\dropout" is perceived
X ko
a
X
o
k
pL =
pL;cum(
k=1
)=
n=k
pL;cond(m) = ( d mm)on
n m
(d: number of drops)
=
P (X k )
L;cum
Xn
E [X ]
(k )
pL;cond(k) = p pL;cum
(k
1)
(state probability)
P (X = k )
P (X = m)
(state probability)
P (X kjX k 1)
(state transition prob. p k k )
P (X = mjX = m)
(state transition prob. pmm)
1)( )
a arrivals
a!1
gk = P1ok o
P (Y = k)
n=1 n
P
1
1
X
ko
d
k
g = P1 o = Pk1=1 o = kgk
E [Y ]
k=1 k
k=1 k k=1
X=0
X=1
P(X=1|X=1)
p01
Trace synthesis: performance assessment of hop-by-hop loss control and end-toend loss recovery
10
10
k1
P(X=1|X=1)
P(Y=k)
(1P(X=1|X=1))
10
10
10
P(Y=k) (measured)
10
10
-6
10 10 00
10
15
20
25
30
length of loss burst k
35
40
45
50
50
11
10
1
1
10
P(X=1)=0.04
P(X=k)
P(X>=k)
P(X=m)
10
10
10
10
-6
10 10
P(X=k)
5
10
15
20
25
30
35
40
45
50
00
50
length of loss burst k, length of loss window m
length of loss burst k, length of loss window m
12
0.9
0.8
0.7
0.6
0.5
0.4
P(X>=k|X>=k1)
0.3
10
15
20
25
30
length of loss burst k
35
40
45
50
13
P(X=1|X=1) = 0.79
10
P(X=1|X=1)k1(1P(X=1|X=1))
P(Y=k)
10
P(Y=k) (measured)
10
10
10
-5
10 0
0
10
15
20
length of loss burst k
25
30
35
35
14
10
P(X=m)
P(X=k)
P(X>=k)
P(X=m)
10
P(X>=k)
2
10
P(X=k)
3
10
10
-5
10
0
0
10
15
20
25
length of loss burst k, length of loss window m
35
35
length of loss burst k, length of loss window m
30
15
10
10
(1P(X=1|X=1))
10
P(X>=k)
10
P(X=k)
10
-5
10
k1
P(X=1|X=1)
P(Y=k)
P(X=k)
P(X>=k)
8
10
12
length of loss burst k
14
16
18
20
20
16
10
P(Y=k)
P(X=k)
P(X>=k)
P(X=m)
10
10
P(X>=k)
10
P(Y=k)
10
P(X=k)
-5
10
80 length of loss burst k, length of loss window m140
80
90
100
110
120
length of loss burst k, length of loss window m
130
140
17
18