Exam
□ Wireless network utilit
– Configuration f: indepe
□ Channel assignments in
– Configuration f: one co
channel assignments
□ Path selection and flow
– Configuration f: one co
selected paths
□ Peering in Peer‐to‐Pee
mples
ty maximization New
endent set perspective
n WiFi networks
ombination of
w control New
perspective
ombination of
and new
er systems… solutions
26
Channel Assignmen
– 3 WiFi channels available
– N access points: each choo
– Channel‐configuration affe
□ Goal: assign channels distri
nt in WiFi Networks
oses one channel
ects Interference
ibutedly for good performance
27
Challenges That Mak
□ The number of configu
– Example: 8 APs, 3 chan
□ Assignment needs to c
ke the Problem Open
urations are exponential
nnels, 38 = 6561
consider traffic demand
28
Problem Fo
maxf ∈F XN
i=
□ Channel configuration f =[
□ fi: the channel used by AP
□ Rif : downlink throughp
under configuration f
□ Ui (): utility function to
ormulation
XN ³ ´
Ui Rif
=1
[f1, f2, …, fN]
P i
put observed by AP I
o guarantee fairness
29
Markov App
maxp≥0 X XN ³
s.t. pf Ui R
fX∈F i=1
pf = 1.
f ∈F
□ Product form solution:
pf∗ ³ PN U
exp β
i=1
=
C
proximation
´ 1 X pf log pf
Rif − β
f ∈F
:
³ ´´
Ui Rif , ∀f ∈ F .
30
Distributed Mark
□ Only allow transitions in
channel
□ One transition rate desig
³ ³ ´´
β PN Rif
exp Ui
i=1
C qf,f 0 =
à XN ³ ´!
exp −β Ui Rif
i=1
□ Recent general results: c
replace the global inform
kov Chain Design
nvolve one AP chaning its
gn: ³ ³ ´´
β Rif
exp PN Ui 0
i=1
= C qf 0,f
! Ã XN ³ ´!
β Rif
exp Ui 0
i=1
can use local estimate to
mation
31
Distributed Markov C
□ Initially each AP randomly
□ Each AP counts down with
variable with meanÃ
XN
exp β U
i=1
– Our recent general results:
to replace the global inform
□ Count‐down expires
– Randomly hop to a differen
– Reinitiate another count‐do
Chain Implementation
picks a channel
an exponential random
³ ´!
Ui Rif
: robust to using local estimate
mation
nt channel
own
32
Simulatio
□ Eight APs, random netw
– 3 channels available
– Each AP on average has 3
– ∆ T: aggregate throughp
– ∆ U: aggregate utility ga
– β: 10
on Results
works (10 instances)
3 neighbors
put gap
ap
33
Conclusions an
Combinatorial network
optimization
• Top‐down approach
• Combinatorial problems
• Markov approximation
for designing distribution
solutions
nd Future Work
Combinatorial network
problems
Formulation
Log‐sum‐exp
approximation
Distributed Monte Carlo
Markov Chain
34
Conclusions an
□ Convergence (mixing) time
– Connections to statistical p
□ Alternative Markov chain d
– Alternative parameters des
– How about non‐time‐rever
□ Alternative approximation?
– May lead to a different set
□ Applications
– Multipath/P2P routing [ano
paper]
– P2P VoD topology control [
nd Future Work
e, and applications
physics (Glauber dynamics, etc.)
designs
sign
rsible Markov chain
?
of design framework
other example in INFOCOM 10
[submitted]
35
Minghua Chen :: http://www.ie.cuhk
Thank you
k.edu.hk/~mhchen/
36