Seminar ppt of a maxflow problem
Transcript of Seminar ppt of a maxflow problem
-
8/16/2019 Seminar ppt of a maxflow problem
1/32
An efcient distributed max-ow
algorithm or Wireless SensorNetworks
Niveditha Zacaria
R7042 1
-
8/16/2019 Seminar ppt of a maxflow problem
2/32
Wireless Sensor Networks
(WSNs):Highl distri!"ted networks o# s$all%lightweight wireless nodes%&e'loed in large n"$!ers%onitors the environ$ent or sste$ !$eas"ring 'hsical 'ara$eters s"ch as
te$'erat"re% 'ress"re% h"$idit
*ach node:
sensing + 'rocessing + co$$"nication 2
-
8/16/2019 Seminar ppt of a maxflow problem
3/32
,or$al -. ,/W ro!le$
ra'h G=(V,E) 3 a ow network ○ &irected% each edge has capacity c(u,v ) 0
○ wo s'ecial vertices: source s, and sink t ○ ,or an other verte5 v, there is a 'ath s→…→v →…→t
Flow 3 a #"nction f : V × V → R○ Capacity constraint : ,or all u, v ∈ V : f (u,v ) c(u,v )
○ Skew symmetry : ,or all u, v ∈ V : f (u,v ) 6 3f (v,u)○ Flow conservation: ,or all u ∈ V 3 s, t 8:
, or ( , ) ( , ) 0
( , ) ( , ) 0
v V
v V
f u v f u V
f v u f V u
∈
∈
= =
= =
∑
∑2/5
2/15
5/14
4/19
3/3
s t 0/9
a
b
9
-
8/16/2019 Seminar ppt of a maxflow problem
4/32
4
a5 ,low (contd)
We want to nd a ;ow o# $a5i$"$val"e #ro$ the so"rce to the sink
&enoted !
-
8/16/2019 Seminar ppt of a maxflow problem
5/32
N!A!"N # ASS$%&!"NS
@ :3 sensorsWireless Sensor Network 6 (@%*)
* :3 links (directed and #"ll d"'le5)
/ink 'resent:3 neigh!oring nodes eg:(i%A)* *
1 *ach sensor can !e "niB"el identied Chas atleast 1 trans$itter and 1 receiver
2 - $echanis$ e5ists #or neigh!ordiscovering
9 here is a 'ro'er deliver o# 'ackets #ro$ a
node to its oneDho' neigh!or E
-
8/16/2019 Seminar ppt of a maxflow problem
6/32
'esidual Network() *
Network consisting o# links that have'ositive resid"al ca'acities
Resid"al ca'acit D $a5i$"$additional ;ow that can !e sent #ro$node i to node A
Resid"al F"%vG 6 c F"%vG 3 # F"%vG
-
8/16/2019 Seminar ppt of a maxflow problem
7/32
+eight unction(h*
h : @ DIN is a height #"nctionhFsG 6 n
hFt G 6 0hF"G J6 hFvG + 1
-lgorith$ '"shes ;ow downhill% whichis #ro$ the higher node to the lowernode
7
-
8/16/2019 Seminar ppt of a maxflow problem
8/32
)eneric push-relabel
algorithm )&'
"nput:- network F@G 6 * stored as an adAacen
listutput: a5D;ow # : * DI NK
# Fi% AG 6 c F i% A G 3 resid"al F i% A G
L
-
8/16/2019 Seminar ppt of a maxflow problem
9/32
Algorithm ,
1 hFso"rceG 6 nK2 hFsinkG 6 0K9 ,or ( all " * @ 3 s%t8 )4 hFvG 6 0KE ,or ( all ("%v) * * ) resid"alF"%vG 6 cF"%vGK7 ,or ( all links e o"tgoing #ro$ so"rce )L sat"rate e ;= While (v is the active node% v * @) 10. i# ( (v%") is down hill )11. '"sh $in e5cess FvG% resid"al Fv%"G 8 "nits #ro
v to "K12. "'date resid"al F"%vG and resid"al Fv%"G =
-
8/16/2019 Seminar ppt of a maxflow problem
10/32
Asynchronous Algorithm
he $ain !od o# the 'ro'osedalgorith$ which e5ec"tes in sensor
nodesn'"t: /ocal in#or$ation o# callernode% s"ch as role and e5cess as wellas oneDho' in#or$ation s"ch as ;owval"es and resid"al ca'acities #oreach neigh!or
"t'"t: a5D;ow or nearDo'ti$al ;ow
10
-
8/16/2019 Seminar ppt of a maxflow problem
11/32
Algorithm
1 # : * DI N% # Fi% AG 6 cFi% AG 3resid"alFi% AG
2 nitialiOe()K9 i#( role v 6 SPRQ*)
4 e5cess v 6 s"$ v (resid"al vF"G)K
E "sh()K while (tr"e)7 listen()K
11
-
8/16/2019 Seminar ppt of a maxflow problem
12/32
Algorithm .
n'"t: he local in#or$ation o# node v which are
the ca'acities o# inco$ing links cFv% "G% role v
* SPRQ*K SNK R*/-8 and the totaln"$!er o# the nodes (n 6
-
8/16/2019 Seminar ppt of a maxflow problem
13/32
"nitiali/e(*
1 *5cess v 6 0K
2 h v 6 init T height(role v)K
9 #or (all " * neigh!ors)4 i# (("% v)is an inco$ing link into v in
) resid"al v F"G 6 0K
E else i# ((v% ") is an o"tgoing link#ro$ v in ) resid"al v F"G 6 cFv% "GK
7 h vF"G 6 init T height(role ")K19
-
8/16/2019 Seminar ppt of a maxflow problem
14/32
init 0 height(k*
"nput 1 role k * (SPRQ*% SN% R*/-) andthe total n"$!er o# the nodes (n 6
-
8/16/2019 Seminar ppt of a maxflow problem
15/32
Algorithm 2
"nput1 /ocal varia!les s"ch as the resid"al
ca'acities o# the neigh!ors% role%height and e5cess We ass"$e thatthe algorith$ r"ns at node v
utput1 - PSH $essage carring a ;ow val"e
1E
-
8/16/2019 Seminar ppt of a maxflow problem
16/32
3 !he push operation 4 push(*
1 i# (role v U6 SN)
2 while (e5cess v I0) CC (" * neigh!ors v s"ch that (resid"al v F"G
I0) CC (h vF"G J h v))
9 V 6 $in(e5cess v %resid"al vF"G)K
4 e5cess v 6 e5cess v D VKE resid"al vF"G 6 resid"al vF"G D VK
send (PSH% V) to "K
1
-
8/16/2019 Seminar ppt of a maxflow problem
17/32
Algorithm 5
"nput1 /ocal varia!les s"ch as theresid"al ca'acities o# the neigh!ors%
role and height
utput1 So$e H*H $essagescarring the new height val"e o# thecaller node
17
-
8/16/2019 Seminar ppt of a maxflow problem
18/32
!he relabel operation 4relabel(*
1 i# ( role v * SPRQ* )CC ( role v * SN )2 check Tcanceling()K
9 h 6 $in(" * neigh!ors v %resid"al v F"GI0) hvF"G8+1K4 h v 6 hK
E #or (all " * neigh!ors v)
Send(H*H% h v) to "K7 '"sh ( )K
1L
-
8/16/2019 Seminar ppt of a maxflow problem
19/32
Algorithm 6
shows the reaction o# nodes "'on receivingPSH and H*H $essages
When a node v receives a H*H $essage
#ro$ node w% "'dates the height o# w in itsneigh!ors ta!le (h vFwG)n the other hand% when one node receives a
PSH $essage% acce'ts the e5cess ;ow i# it can%otherwise% it
sends !ack a R*,PS* $essage to the sender-#ter that% the sender roll!acks the '"sh i$'act
and recovers the 'revio"s state
1=
-
8/16/2019 Seminar ppt of a maxflow problem
20/32
Nodes reaction upon recei7ing messages rom the network 4listen(*
"nput1 -n inco$ing $essage #ro$ the
channel with w as its sender and V as itsdata oreover% the node "ses localvaria!les s"ch as resid"al ca'acities%e5cess val"e and the height
utput1 &e'ends on the inco$ing
$essage and the state o# the node t$ight !e an "'date in local varia!les or
calling other o'erations to 'ro'agate datathro" h new $essa es 20
-
8/16/2019 Seminar ppt of a maxflow problem
21/32
n receiving PSH $essage :
1 i#(hvFwG Ihv)
2 resid"alvFwG 6 resid"alvFwG + VK9 e5cessv 6 e5cessv + VK
4 i#(rolev U6SN)
E '"sh()K i#(e5cessv I 0)
7 rela!el()K
L else send (R*,PS*% V) to wK 21
-
8/16/2019 Seminar ppt of a maxflow problem
22/32
n receiving R*,PS* $essage :
1 e5cessv 6 e5cessv + VK
2 resid"alvFw G 6 resid"alvFwG + VK9 "sh()K4 i# (e5cessv I 0)
rela!el()
n receiving H*H $essage :
E hvFwG 6 VK 22
-
8/16/2019 Seminar ppt of a maxflow problem
23/32
Algorithm 9
e$'oral Qanceling
- techniB"e to discover tra''ed ;ows and redirect the$!ack into the so"rce
o i$'le$ent s"ch a canceling% each node $"st !e eB"i''ed! a ti$er and each PSH $essage $"st !e a"g$ented witha ti$esta$'
Qonves an idDseB"ence that indicates the loo'D#ree 'ath #ro$
the so"rce to the c"rrent node
"nput1 /ocal varia!les s"ch as the resid"al ca'acities o# theneigh!ors% role% height and e5cess We ass"$e that thealgorith$ r"ns at node v
29
: li ti h k
-
8/16/2019 Seminar ppt of a maxflow problem
24/32
:anceling operation - check0canceling(*
1 isra''ed 6 e5a$inera''ed,low(PSHX)K2 i# (isra''ed CC rolev U6SPRQ*)
9 while (there is an "nhandled PSH
$essage containing tra''ed ;ow)4 V 6 the #low val"e o# the PSH $essageKE " 6 re$ove the last id #ro$ the idT seBK e5cessv 6 e5cessv D VK
7 send (Q-NQ*/% V% idT seB) to "K
24
-
8/16/2019 Seminar ppt of a maxflow problem
25/32
Algorithm ;
Nodes reaction "'on receiving the Q-NQ*/ $essage
"nput1 -n inco$ing $essage #ro$ the channel with
w as its sender and V as its data oreover% thenode "ses local varia!les s"ch as resid"al
ca'acities% e5cess val"e and the height utput: So$e changes in local varia!les and
another Q-NQ*/ $essage
1 i# (rolev U6SPRQ*)2 resid"alvFwG 6 resid"alvFwG + VK
9 " 6 re$ove the last id #ro$ the idT seBK4 send (Q-NQ*/% V% idT seB) to "K
2E
-
8/16/2019 Seminar ppt of a maxflow problem
26/32
!he adapti7e algorithm
n this section% two new o'erations%
'"ll and '"sh!ack are introd"ced%which are "sed to ada't the $a5D;ow in the
network when ca'acities change
Whenever the ca'acit o# the link ("% v)increases% " invokes '"ll o'eration to
trans$it $ore ;ow% i# 'ossi!le
Si$ilarl% once the ca'acit o# link ("% v)decreases% it triggers the '"sh!acko'eration 2
-
8/16/2019 Seminar ppt of a maxflow problem
27/32
Algorithm
-
8/16/2019 Seminar ppt of a maxflow problem
28/32
Algorithm ,=
n'"t: /ocal varia!les and dv" which is the
a$o"nt o# red"ction in cFv% "G
"t'"t:
- PSH>-Q $essage carring twoval"es% one is #or ;ow diYerence andthe other is #or the height val"e o#the node
2L
-
8/16/2019 Seminar ppt of a maxflow problem
29/32
!he pushback operation 4pushback( *
1 i# (resid"alvF"G I6 dv%")
2 resid"alvF"G 6 resid"alvF"G 3 dv%"K
9 else4 5 6 dv%" D resid"alvF"GK
E hv 6 init T height(v)K
send (PSH>-Q% 5% hv) to "K
7 resid"alvF"G 6 0KL e5cessv 6 5K
= '"sh( )K
2=
-
8/16/2019 Seminar ppt of a maxflow problem
30/32
>rawbacks
1 ost a''lications o# WSN are sensitive to long delas !"t
still the 'ro'osed algorith$ is slow in so$e sit"ation
2 he algorith$ does not have an a''ro'riate ter$inationdetection $echanis$
ie the rela nodes and es'eciall the so"rce do not know
when the algorith$ ter$inates and what are the ;ow'aths
,ort"natel% the sink node has this in#or$ation and itcan in#or$ other nodes ! "sing the idDseB"ence
90
-
8/16/2019 Seminar ppt of a maxflow problem
31/32
:onclusion
n this 'a'er the $a5D;ow 'ro!le$ has!een restated in a wa that the o!Aectiveis not onl nding the o'ti$al $a5D;ow
!"t also red"cing $essage and ti$eoverheads and !eing ada'tive to networkchanges
he algorith$ consists o# severalhe"ristics which $ake it asnchrono"s%lightDweight% ada'tive and realDti$e
91
-
8/16/2019 Seminar ppt of a maxflow problem
32/32
'eerences
-n ecient distri!"ted $a5D;ow algorith$ #orWireless Sensor Networks :Sa$an Ho$ao"nneAad %-lireOa >agheri C 201E *lsevier /td
old!erg -@% arAan R* - new a''roach to the$a5i$"$D;ow 'ro!le$ -Q
-h"Aa R% agnanti /% rlin [> Network ;ows:theor% algorith$s% and a''lications *ngland:earson *d"cation /i$itedK 2014