> no$7{?NPNG
IHDRddG(qqrrsJ,tdUvusyS/~}V/_:mL~a`uc\s0&MdLKS+GY&H!b̓'/ƃR(g^ũOXk$Xp?*~T\rGB*@W,ɰuZ[(_9aH[P
9n|J3ɘaQ\C@x^H|
VK
@ai8!&J:4.IY}%,#%0=cive0|sӪml,m֫V9i+ɡS
ʺhD_
)&^4>7g 21 IS(ap.DY!7
d"m*>(V( >3p}q-
T[T,
5BR(b\'RdirO:Lkh):l8(Gnb_UJ6,<f6ey(Q$0ȥRT 4sJ8dq3 5rY*)<"K]#O
v/yC,'
CdQ{CϤ!] i6ن[AcTXpa'HbC&cǤ#T
)ƥ59GaS)DV>\ţ= G:>p٣^!4D2"y%&>> 1dc
ŀppLJnF]VCzK{MEA(ʇku:A@-,hש7>i|V+c/@KT^|
o\Zu\X_`+dmq8~2`l/7N-;.>]<2Z
BH"fRٿVWoC
*5N[[ iRF7J|3l^ǋj^pg66\R]i
=Zz=w-nDɨ,$+E$jۭvOV[ꩄϯ;
a8^5N+?VCԝnkElێ]t9n>w.sZ`G ʊoaO\fG{Wou{#?v9sG&df^7'0w2$#]<ē̢(aj%˟d8~9$e$?k yyҽ{|B,3<0Q\#bGB|(b "ع[_.B|ϸz;\/VF"'Swb8GYp4rDsl3Oan>ICM_~`p
No1$N!K,\-d[P)yqϤab\4BX߀ۥ9E߁xV?Cף?6I߅\Ixd>_rSamsOGMSOFFICE9.0GIF89add!wa) :$$P22g3!<(qqrrsJ,tdUvusyS/~}V/_:mL~a P]__QjT_Uyق@AAm_R`F!@M>.0hvł Q$@D{be%qq|0 $D
Iǽk^%%LAίj&(8QxJ.ԥ ~*_JW"fERqorNЃ&QQ,[H [LZ*]*:q=ժGո^5'J
RC%"@X5p]* k^3)mR/#b5ABuxd֪se#aYpl2: Y 5e*Vx*e@քL&A|u!6!W*q- ^0M 4$LH< tFCrI*10,/SH@dꆳ[$#WdԚ
\@DmT0($8?4E[XP-8K
7c bH(wWbX &%DU\χ>J @A
t@@Qb
P@bB`!V2ZO2`O}xP@'/-(DVp0"ݬ>&тJ.0 1y
Z2Adhvp\iI(`؈&V|g~2
p@m@65#:p٫
7;ML:>^@()~S$;߱S<$`%(ɱ@5}'v&уykCt٥Jh8Ђ&x
TI[=>ЁYW%h@j="pN_}if=`:OABX;vk BWwBP@s^IpDvC5X@+?" "PͩsqǋO@xK:O>A
2>B
$\+
^d`?=GZЀ _S.&3* u|: "@ @׀ 6{(r76x/}'w+gr/Ȁ`w!h:s?0"85hs+}p7y)@58s6/Xx%3^0)+0@h()uIF{188Y@)+"P/@(/|٧}P}5;g%-0r&/ye*) Ǉ} `9A`
<2=0-&~8 ?%( O(#ب)# 0o[# 긎91gx -&3`xe荁;[K-Ȏ.`)Y{X%ᨄ"2|٘ۈ2`^(~4 .-Draxt >Ў9-ЁX&!z4Aސ
Q
^ /)3
iJ(]ؖ{h\IoX8|i 1p 9`X7{`Y،8%xa3Ћg陨ٚ!
B, V[@@;96339=3;?3.1=9.=1.=61?96ʒ6θ?9ͬ;& "6;?;6..&!!)6=1%G/Lτrp,Vl0hU!/q$FA$+!ĘeP%o"fڃ4ç%VMC+2k@C ΜrO)0k7X4Ղ#q
E3
zU
`@EҜpÇ
-1J&Aq.686+v 0`B\)\=
FAcE7L 5(PpܲDD4@н`A( 3q)AlYJ
TCtPh(xl. }=CTpb}
HXd-"@
Hd':#BIc}[;°x@ EXA/ l@jV)XeD
>eh4g.(p7`jBhBV,`
YhA8`g/P&U^Hcl)xEڤ x
ں if:,PE`룵"
lݖjfbF{@3@hz-qtzkf0+4
6:Y4V`@0t)瑉ʼ /z G9;j
pJ$J}sU[昇Z
wEh6)`a{F^}o;gKJ$2>-dv@ʝw範i#
9>H $ 9碸P@J3PNYZg9s 륑껤\@ش7ZD#28&7A/Y\×CAt6Dtw@ԆT9կPM2 f$d 2
@$ (,AD`kJ`a1j V" zh
QiB^\w$nrR HdpЕI@HX F:)hKT*C/R
ph!ǥAhxF$63)*UV@nqzVIN2eiSؽю.$BEld.,_+P
;k@!v'~`ƗPi"V%=Q_r"@WL\&aIÑ&P,c,p**e*Ӌwn@q
10ș4 h0i0O5(dR%L 孠=:kB /ExrEo[k!Rk6:xVFN$~yf%u
zti,Ē@\GDQlj1AkFВ5JTb(8]pЮti0V>APL Ώ
`~@/o %d;f}%s\`;r_^P9f)#D d9A+̓Po|yYb)TmW-!}.DAV:eރȁ^w]HW`/:B@Aw:수̽͝ko :hm$a8nu@"Q@t_<?l "x)"1#@o uc $ޔ9E h>=loRp-p[_o@s@ ==/#8TJ#*/)"@WXevjgwGxX;0{
?"~{8((,x]z)0yg5u#:X7 3$hQr+(
0k;hG1{s6u+P&')0%( N'yX7-\X@~-g|g:Xep2H]gu*3`[Dgb; &`+aȄ.=0{90(3xvw(`- ;DG{`"0&-Po
!H"]V#;Af8@;@(H
;Pn#}w`1-PA08AƐs1g4`tXzk"`ّF)wX
8.IAX0XiF:P#}6I
6\Ey6HI Mi ?`Py%@SyGyP[I ?`_3
Y`!
B,U[@?9661.6=;36.1;91?=..@9=931?;;ŉ?3.;69ׅ11).%13Ӄ;.!3ʂ.3!r%m.|3/ @)msBuBC%9+@@y~i4ÄRpq i`R +^/`#48"g> '
oh[+ IŨ{
,\[y,=t
0o.*B*A&A*Dc4P@7+t$ bHfJgz'}t;q\BK\p@L0@:Zs-*Ь),K ;@ŉZͰp
y{{n "#~ҋ
y H&@ 0P9DׁcL7W6l]6ȐHw+H0c@.qcUf]NX:AD~nQ
( (p2+_@P6Dn;U`]Z.X-v4;={R$
p:pYn'Ba*\nvDTeH
\F
7<@2@V4eLO>RzxIİRxJp+`&gŋfSPHM}TҕM;~B1KJq@_ɂ-)v)YԀD4
bk
lc@GTQݥzFi,H.I0'q`$W\<8;Av:-v{R0PWVʳi<!lIt
ҥf{\ ĂNNQ@*RQq($B($v[ '554ʤ
̞J4',z,L@8f_[ǺLM$` 4SpBؠ?+_UຯAV*AebWUx
AJy@\jA=V94d1#!LHˮ
V@y{hM0ӪV%kE+X v]!*8o$V4Tw&aTu> - A"x(A/"
L[KD!(7-CF7]<ӈև܀^@A , ͠Pq,$Oc`'`*E
H3Xw?G8rsܧ|b3$0 * `pV3@11-p)I6<)HL,
*e (83!k˴<B4t4Il'Y>8F4-.VJ [60A(0:G\R/]YBm؈n[t`7(@|3#9&ȩw;(
R\`mL^H@9uR|mEQܠ} {ϑ^L[қ
ts/95у1_@[kd?PLDϺ}PM_eq}
5(u
n0nM>ZLVz!H@h'
CO{̿]g@ yn1@/|z@D`B0^/ׁd`% .!D# OTP1p J 0[40t\)8hO+7?007LtBq0u3W/%@װ10*0qW++
0x+~{v5\41 78"2hx/ߦ3nׂ?`{U'x"+0:@h&0Hx+qKȄ׀P%p{PX-0M`"H"0p@-p `z.xy7PhO/hUg~8!{\]*
qOHv&
Kqx"
0@§Ұ)w'xv×#%Ќ.``"}h)p&f|͑/zzݷ
⧈s 1g"Pa&x'wyǘw}#`@J%P@X=
zyG|#-pXS
%+I|+pp#5)Px%$:iz( ||+ 1ȏ41-0#[ؓD7
T{|Ø9g{A3uw̠|Rx9YI!
A,UX@?;9.)-.6@9#
-?=.,((408488+
.;
044
&ِ)!
BAyvpB|kC|
!2:TW7=)r:Ĳ/E]7?/(hX@[ff`~(9L(pv`~!@Xq @
m8i7scc9$"0@PA;Jb(\{+4@__*IqrHJr,/Sto0&BHe@M(#&^Ly3ʑ.0@ܬB5DQjZb~iZGVo\4A#CFhsb[aV+/+H5{!8VdK*%>Ĉ7OIO>x
MYʡܚP%,%Y6b +H
u%f5t0wl"˘PMؕ :j@A
p0L*mKc"À%;Z
$Kl|;MP/"xrM &;4bk+r`'.}ILt!,4t4s
IUP8^VZy]#x-ac,`8v)
^PWF{cBTw#SӀA
;q0۫FxZ]kΊrP@@2;$@H
H\ h!vg&gA}k^<6j.A;>!=|QC5P22~t;$ 'k9-(qlr
\Myw߬-@v}W(G ۺ6@$(fy/83~|w*(CO*_y
4k9" 3Ȣ0lT+;W3(4L=6 0!߹n W/*x}\L7@W#Xg
x2ЁwB>^"Ȁ X@P'/zW7'z 2_GjzMLǻu A훎w>)"
;wnG
b IĠ"ɪeu@}#s|WG2"#G})&
ߗ~ g"qq̷~e
91 |6p9i;p9&}|W m;W&0.8+h)0§
07)
H2x6.`"tQWɆ2(-Ѕ- EoSP#P1t3P`Xt
>Hle+橭D@`Of"QB8(W*^/8*+j2݊RQ L8QT:N+%v=Ad ,aj±@M8]N,Ę=Jx=Aޛ,WPa#tQ2
^`*% )
-Q*)}#"9ˉRX^*1 9~Rp?"<4c:THA*w}Ek8}:H4veՍv%]PpĴ$!),
fo}+pp$tt@
942+gU]TA ;
ёA4KRkZAk7ɱtF`nlTKFMV% J;.@nŝYJZ7XnxGm0:hʰ&55cdR"XZ&!
P74KDSP!~p`-ZL>+`V0*r8%$\w`PF˳`MoM
"IhQݡop$x-OJؠimV
4g],Cx9T$Vb4hA%S-|K%Meh n4Hr4ВXSu#ajWZ]+H*p4A
|m`Gyug
'gBsdlW1
-)nvL8MC 2d5%:q}jՐg&B6@tM:Es n0D"Aȹsn׀NZP`&29%f`A R;rI#09(Z4]O7ɹ=ug@)ADu1H@W)@0A{ޞ5T@# d{Rӽ>
^0q^2 \:y|۠_ EM@ O}@
_{纗
@=PDǿ{2-+4@"cP~pw{')0P wzq's{$s~$8z4h I!0!|sg@4zG"Ho4Jq9cV1Pp;b}>&l0Q8 $bxX{2~*Q&"p-uGхXA-0 Q:*&h4H%pG(q1ghQF
0應**P~[90f=@$80Q(ڸH8 ypy787W(IwXG
-0@"0(M`9pi
+ .vXxwXna8'蘉?7~3PQ(倒H%`X7h;4hRE0H%Ў;P
@Q|o'
JٔQ&pTi?WS_ D) YW6h9c.- u1qi!
C,
VZ@?996;9;?@=3.=1;1..9=.@693³??6=̑9391631܅)-)е1;!岥6!ل UnU+? 01RšDxq
t[SEqda=.X.J)fЀJ;TcDY&v
(#ǋ^dCQ+18l@VH.CVǝ\!
L=w $&^U0cvHqDd5}U+SlbƊ 0HhB.4?x a (p*va)W.J(pܾN|`)G5Ar`Eo` |lA*@E`U%hhuT@keXA"6 Fax !
@h+T@bBv+ DAd4a`
Q ;|{bo>i< ( &eXHPk:qFpCJ*,nڨeA~ة He]75@ZY,'@ b~ykع&RSY&~* nf4' a؉W`'c۫*X ?uU~ nrLfx2m`}.)8^Akk***펬 f*oمw+hjBj$Z/eB,3vsYjj|yP Q%&=`تvdjywT KI.p0^@H>:^'C(wjxlC|z p |`Z_/[A{K__X:@
~ؠ0^+{B=Dȣ
10>2XAХ E,9qӝO`@w;<^&)PNm"& 5S(CvثY
b(TV2|A&dnM[V9se
#`m,ҁt8epQߣ}B̟Y["!Tz654 >klRV=ro
HM`~FS2[fMƂiH(b r§v0zӓ#dh& b`9 C<
xB0:Dm^in>*@N)>S("pPl̃agԨdS@I*UDa!f2P xj%eiV.B`=$@BcH=e,֧qJ`ެT>bPiQWE"j W{CUv%tz,iтoti̯o&>$pp6C@45>
&5
ίT g1@ykr70!bpxAoZG@x(KH<2(0j7AWj7xF0(Ո()BPb14 _\2
tP@Xes'т>ꠑ>l1)q[)9 ~ԛԹF -^ .@P9pߴVkd'Z
\\#((5X%$燬x 0w;qVvztH j@ފu}d `MOt]Loj\6T:f/}{; 7㮃$(9P~ȯZ|bk~ζl2ϛ"i"&>,8W_@*P:7lh?Fh-
}1?}!@le;D[ۙ0
}z2i@r 'mږ*P+[7X~Uzk@ zmvPhAuYG%+u[ #H-`-p*Z8} `7}'"p#8 0
WPDX9X""r?P7=PŰӐ[Ȅ `V7*h.p@
ox"a3(APA)#U**"P\-P#Xt88-7 P
;(A#p=p `t(`~?8"n8HiG¸A]1` 9A1
O؋&Xx`s.%qɐtxՐyBhr%0q בKQ,,9|&
89Y9 u0TCn)Ⅿ* Y,|ᡉ*C/*X ,
I¤qF'i1$f2X"x$XBdF76j :3~,n%;yH4rj!DހX:
h@1[AJ>3q%Vxq,`5Ct
+] h9nc D+10 9\
xj8=/8y6%6p,O^; 64gI
@&N3Tpb<|*$h?"Ln#_qhwl&xNk
hkA,@ @xA>7$DE9)n- 0 c7P +Qt4!-N{xTbc;j{FB
S!XR.f4U8c2.RPcS
Q Ji,ХofRK>鱆jZ XACID
P;AyڋmgCR.AABV#x>F
Xh1D2HFLK
u(р@2t x&m8LOWr Xzf'NL~:~ [%iNUDRSԴ @7Wub(HkXqvQMe}
0TP^L*R~H#)W*NYS4ms4MZ%ʺ?r(""Mr\֭g,
)^2`0ջYw 4{fSȬk4\ZW^Յ:+HyQc]g;I0[nXЀȖfO@[DC91|Y-+W莅A!( 0wa<]z1aepB`f̞<@Id5dpMqq|gΤn, N7C0,(At!
}Ε.D =!fB)
̈c>:KiʋVU'&[Eb$|M,zĮK*2!:t1!xP ]Q\I2 +%G {j7}H/ 0x<`4p40@s(i4yD'@&]S,9>@LYL8e !v7wӘm>*P+op27
P`~
4G'xB8Ch# LHRsShzwyWy70bЅ'8?8b(+@x7*0s0Rx)`].03R~z/Pp$#`AeX~5Vۤ&KAs"|Gw7=;;.p-`~1p39
)8@;%@Xޣ|/$3e"{bgXʸN'X9*~淎Uzp818X@=;~9{'}:v02 ل%!'
|Linxv>< <A`
Dya
#` X 0--Ж8!uz|ٗ!
A,VZ@=963.6;?63396;;1..9?1331.1??=҆.)-"1;9&36.!)9
V
}CDnŸDՎ[ W#g+e;l+[.qp)r8Eӎ+nI-)C\qC-9K8uA6~ a&VC#$-.w*2w"Nh!tXA-
9et)`5L74`5Nͅ
meh c0
n@k&rD0DkJS$P4rnI 8E0`xݴiW^w̝@3Xe@f|A wȈ@xPA t"(
X
4 _{PB @A
H &0y z~V
8Eb $6($ 0`О}HdP:$c]T "DufIB@
}FIdDD
#_`_fD**b6pzk9)$*BE/B!V"*xAT4BӤ+z"bhvSfABV㹰wgaÂ*F~ i ,F Aﾾb@C闂 A8гC~~Io+Jc ptg))=t HmPg2BZLʹ@0
/Mty"^'z@@8s_Z
8,.kx)ꁶG=@@ܠ&~Ħ_0py5}1$
y?#d7D
/ɶ
}Dcu"Bt5(' Wz{*\ߜHu!}t3 As(A"@܁OXU!QP*D$+PRj
KǾ%m@!r\R@OZ}H.Re0PR
[a#-sKd1XB赸****2._HX]0h'GG(:z
Pt8
-`7;t@@0u.Չ:8h]uL+#~\Jӌp0m[
xP+tjj`%45A+{VNpk58btPr> 4A@WekL?(6j̅:8bTȀNxZvWHf\UFX=ܠ[ '|(OZK^\3x] P#8x
캮|
@0=:8>yѝމP:A`a;UN22X!P?-v (vv`#T݊!
V~0'x ܇/b^@DʹO^'xWt@r^]Qh!L@)(YO@*к4v7;` h -
QY}7}L
(@zz]~Uo7|r5p"0@~ 0'x+|()=0{`'Wi/ G 0}/0Xp 33wv}gm~2(@0'l-CCJu&`0x-}'nXf@ @P 05P} ='>`E@Ch;.~ H~Ǘ"`{0&`70 #`rv=~$ҲM(l1
אPa-~
4%|L
YK{8{И)`X-0&0@%@
)Z/0726 '"`3YC?#6Pn99pnl
&XP!G
U1vXE$8`
4;O( 9%H1$"0H8 1`GFO1 5- {vv0V-pZ;p] EdY8hy\ɖo!C
nYC $d#&0YY;iIENDB`@=Q1VIG
<\0D+\GxYklTE>3Җe4݀[
]X)bB"UlP)I4`X_?DlHLC0[mA9lg}Bq8g93g{E!#41^8 !] CHr,{3
Ktw"n~Ȅ·dyR^ʌwxM)5C+j 84b .e۶5]/9Bڲ\e:5fFpJ:ţM
Pq2In+a/_^AGom(ڲO-j(Yn:2fg|1"]qR:k Dpۋ
hOYf\|KZ=Sȝh3̷zd"[lqZDg#@Gņ!`{uA'v
_8iYҷge=Fb>S=^PuN]2vWpQ惜D ň~SϜ:!ؾܽJx$[&*s*2+ib^'}i)eql Y0=>A|{Ty&/Whݚ[7hM%U_*oJ2+*_cTW僚|P4bTWdoS2=J9չRczN4Λ)P~f[>Peb~-1ÈV;&fEoG3[Xe^B#VF0'=<ꌝ'{su|{f4ꃕ2b~31
_aď!ytg81U+o/x9A ev#B{
}c|kwF6BZ4j3A#ҷMjlYxDv8.-۱5&Gv߭!
^Ě|vL.ՖtNk5&8s2L|߄:wsWR`e`^5kn{3RϽFv+g ݉u$)(V>.I>Snl"HUirj|; bSx
u{\p+EHͻ>1X xko[4xK4^z7"ށCIxy$dO}:lS31
'~!SGw)e85rJKy.c9DpYq6nVN)c)y,twSRqnd2<$|oﬗ)a>l7 Rėu;PxyF3G)|>kh>uf^~/<6IDUKYFpաʽ+=wV0g>';|%o|$-9( 5RJ\Gw?*{EnG6=>ݗE*PNG
IHDR$O&PLTEȝ@_Hɝɝ@_ Hɝ`ȝ4ɝH_ɝ@_@_ Hɝ0ɝ@_S@_@_x{|ɝ5H_`ɝ;6ngtɝ&\~dB˝mnF<Lo7'O7'Ov=@_v=dz=8x8@_8x?@}xxftt8xg-L:sw8xggw8x H* =ʝ
w˝&$0@}dҸfXcfXct˝fXcfXctqbKGDHcmPPJCmp0712HsIDAT8OQ0Di&vڵOkb)ywYcDI6HD]Dt+:`*K,v~*'uZ}ccw'#+%CrD&@SA(Z5V| BZtN$y~l6̄. KmC*u=pxt+X
(@+_[>ԷDIENDB``!!e)e튨hC& p* ,MxڕV pNW>91tNmigU[UQP:)4ڪ5$K-Xc !B"J-QKN?Djܳ}{}/"爔T&;SZR幺?
3HWtPAFg5U)R7VGzg^I+?`?^le
u]8iDg&9 ܇ND[ߥmu ݜ^vz"%7|P DEUE)GlCh2H62"2%d;3\ܒ(0EeB
Em? B|(R>lX/H%Vq%Y̗;B
/)
zbB1G1Hn I>idzF22&E)"J*RiZ^ӵn:^Z}Ut6BDǴuCƴS֗驣W>ÿd
V+`/b3 X@#{q0
G9J:մ/ި=XC@ZOb}j
6#l*F[,F[XIl" h"O~~R"s(SDX#t7u.m1˜&.mSnLoY:T(;1E7+b\s9ۗmӦYKcAԻUZfaLpN8@4TtMxZMyjh O
n.rO/}~D1LN^&qM)d1#89zu4B/p¯ Xo-*9$Frus'A"P[b7ː@klb/Ѽ{vԝA;7/ /" ]9ɷȯ(|W&[D`U"
۬1~fȰVgmeqzCED?+њv$D$KOf7#>EtkʛEg_yWNM!E
!F=}"}]ҊurORSM{4gQL{GמA?iWGCJ;-tkbݕ>f|d{u2zGqy7;yn[Ki+6E-_((gSl20bX֏gK*O_Oz>LֈrnJ&ym$!K,=m\)X'R?B O3,/0o|-"S}QNaF3Fh֏`D5bQqm-9dyǲ˦smr3(
83
Chart MSGraph.Chart.80*Microsoft Graph Chartb/0DArialSans}8 08z[ 00"DComic Sans MS}8 08z[ 00B
B.
@n?" dd@ @@`` X_ 3
_b$$7{?No"$1VIG
<\Yob$6=>ݗE*Ou2$e)e튨hC&)Kz2$iC@@W[Ʋt0AAPf33333@g4DdDdPz[ 0ppp@<4!d!d| 0X}<4BdBd|< 0ʚ;ʚ;<4dddd||- 0h0___PPT10
N___PPT90(?
%,
d
Game Theory and Computer Networks:
a useful combination?
Christos Samaras, ComNet Group, DUTH8e;33%,@
Game Theory:
general theory of rational behavior
What constitutes/characterizes a game:
group of players
(more than 1 decision-maker)
interaction
(interdependence: when a player acts, at least one other player is affected)
strategic
(a player takes this interdependence into account before deciding what action
to take)
rational
(a player chooses her best action)
VZ- --P--]--'-33'f
fPf]f&
consider...
some players that simultaneously transmit data
(a transport entity = a player)
they all share the same network channel
(low bandwidth)
each player chooses a strategy
(e.g. conservative/aggressive behavior)
every player is rational
Let the game begin !
(but... what are the rules of the game?)-3333
338 33w 33f
Visiting again our Internet Game, we have some answers:
WHO& ? a transport protocol (= an Internet player)
WHAT& ? transmit data
(aggressively, conservatively, etc.)
WHEN& ? no turns, no rounds&
(asynchronously)
HOW MUCH& ? gain? lose?
Utility Function
Z933 +33
:%33#F33S
Utility Function (= Payoff Function)
& specifies the PAYOFF to a player for every possible strategy combination that he and the others might pick. (Internet game: a player's payoff depends on his transmission rate and congestion/delay experienced by his packets)
The independent variable in such a function is the player s strategy.
So& what s the best strategy?
dp$33f
Solutions...
(to a game)
D!33
|Game Variations
Extensive Games
... players take turns; the possible actions a player can take on his turn depend on the previous actions taken by himself and the other players
Repeated Games
... a standard game that is played repeatedly
Signaling Games
... the players can signal each other and share some of their intended play, or private information
Bargaining Games
... includes states for making binding offers which the other player(s) can reject or accept
Games with Incomplete Information
... the players have to take actions without having full information on the different factors that influence their utility
$}P333333333.3333e3333^#3333|
Internet Equilibrium
Game: Internet congestion control
Some characteristics of the Internet Game :
repeated games
distributed environment
conditions and number of players change rapidly
Internet equilibrium is
not achieved by rational contemplation
but by interaction and adaptation
...in an environment where conditions (and player populations) change rapidly (and in which changes in strategies incur costs).
7Z.Z]-Z-ZZ7330f f0f336fH33
j
Nash equilibrium & is not (always) the solution to look for
Why?
efficiency, fairness are NOT GUARANTEED by Nash equilibrium
actually, a standard solution concept (like Nash equilibrium)
DOES NOT APPLY to an asynchronous and distributed
environment like the Internet
MORE SOPHISTICATED concepts of equilibria are
more appropriate for the context of the Internet
&H-m33+3
333fCfF
f[
A "perfect" design:
4
A (more or less) tangible goal:
Develop a reasonably faithful game-theoretic model of Internet congestion control for which (an approximation of) TCP/IP is a Nash equilibrium.
To be defined...
PAYOFF to each player (for instance: a player tries more but gains less)
what kind of PENALTY for selfish/greedy players?
WHO is responsible for penalizing a player? and HOW? (network: regulating role?)
desired GOALS:
1. for the network/Internet (better bandwidth utilization?
lower delays provision? fairness above all?)
2. for the players (better throughput/goodput? less packet losses/packet
retransmissions? smooth data transfer?)-V- -334f33R3333-33(33Vr
tSome facts about the Internet Congestion Game
not a "common knowledge" game
asynchronous, repeated games in a distributed environment
payoff function not known (nor the payoff functions of the opponents)
little a priori info (about other players/payoff function(s))
little or no knowledge of the underlying network topology
conditions change constantly
0 <H@<3333$33G
a
what next?
The aforementioned scheme calls for network rules and actions
(which will motivate the behavior of players)
...in this context we have selected TCP Vegas (throughput estimation mechanism)
remarks / ideas for proceeding:
network/Internet should not be vulnerable to greedy users
Nash equilibria are not necessarily achieved (as a result of learning)
nature of learning and convergence in the Internet
reasonable "learning" behavior-<--I--5--!- 0
` 33` Sf3f` 33g` f` www3PP` ZXdbmo` \ғ3y`Ӣ` 3f3ff` 3f3FKf` hk]wwwfܹ` ff>>\`Y{ff` R>&- {p_/̴>?" dd@,|?" dd@ " @ ` n?" dd@ @@``PR @ ` `p>>f(
6ɬ `}
T Click to edit Master title style!
!
08̬ `
RClick to edit Master text styles
Second level
Third level
Fourth level
Fifth level!
S
0Ҭ ^ `
>*
0 ^
@*
0pܬ ^ `
@*H
0h ? 3380___PPT10. Default Designw0(
l"
6"` `r
S/@`
/
H
0h ? 33___PPT10i.7+D=' =
@B +
( @(
Sd/ `/
H
0h ? 33___PPT10i.M+D=' =
@B +
4b( w
4z
4 C/P `-
dB
4
<D| p@ dB
4
<D|`p@`dB
4
<D|pp@pdB
4
<D|
p@
H
40h ? 33y___PPT10Y+D=' =
@B +
'
P2(
S$/ 0/
"B?qq$jqI8g ~q$eq%[qv1cO0T
c$X99?jB
B
S,Ԕ_s5142"
7r r2
#z0-ood_s5141C"f
18
@
WHAT& 33jB
S,Ԕ_s5132"
@ @{2
#z,-ood_s5131C"f
{
HHOW
MUCH&
33jB
S,Ԕ_s5130"
rIr2
#zl/ood_s5129C"f
IP
@
WHEN& 33jB
S,Ԕ_s5128"
@i@2
#zD/ood_s5127C"f
cj
>WHO& 332
#zT/ood_s5126C"f
D
GAME
RULESY8 sn
2sn
0/
soG
Qis playing (group of players)(2
0/ 0
['they are playing (available strategies)((2(
0h/n nO
_+each player gets to play (order of playing),(2,
0/
Nthey stand to gain or lose(2H
0h ? 33___PPT10i.c+9+D=' =
@B +K
bZP(
P
P S/P `/
P
X/GBHqsdafsdf"`p
*
P
0/@P
HIn game theory it is standard to
assume COMMON KNOWLEDGE
about the rules4I(2(33f33R
P
C*Aj0286034pP
H
P0h ?P 33___PPT10i.!D;+D=' =
@B +
p(
S/P `/
X
0X
0
X
0* X
0pXB
0DfԔ!
H
0h ? 33___PPT10i.S\J+D=' =
@B +)
@8D(
D
D c$L- -
D0
NA3? ?nHl!3-`8 A
G
DKiF
D
0w-A
GK
B
Strategy_3(2f2
D
62NH
D0h ? 33___PPT10i. T+D=' =
@B +
L
(
L
L S(z `-
L
0z "`
Dominant Strategy: strategy Si strongly dominates all other strategies of player i if the payoff to Si is strictly greater than the payoff to any other strategy, regardless of which strategy is chosen by the other player(s). (the best strategy)X 233
G
>Fp
L
0z "`O
t
Dominance Solvability: a strategy S1 is dominated by another strategy S2, if the latter does at least as well as S1 against every strategy of the other players, and against some it does strictly better. (try to find an undominated strategy: it s a good choice), 233n
L
0`z "`P
Nash Equilibrium: a player selects the best strategy (which yields him the highest payoff possible) assuming what his opponent s strategy choice will be. A strategy combination (which comprises a strategy choice for each player) is a Nash equilibrium if each player s strategy is a best response against his opponents choices in that combination.,] 233JdB
L
<DԔPdB
L
<DԔPdB
L
<DԔG
H
L0h ? 33___PPT10i. +D=' =
@B +)
@8`
(
`B
s*"`Pz`B
s*"`3Pz`B
s*"`Gq`B
s*"`r PF?
`B
s*"`8J^
Sl%z `-
*H(H
0h ? 33___PPT10i.G+D=' =
@B +
x(
SlFz PP
z
XB
0Df|p XB
@
0D33|` P
H
0h ? 33___PPT10i. B;+D=' =
@B +
(j(
(
( SX- `0z
RB
(
s*Do RB
(
s*Do
(
3BCjDELF$ԔFh.-F\o0{Ji,dOj@ "`@mdB
(@
<DԔ
B dB
(
<DԔH
(0h ? 33___PPT10i.@A+D=' =
@B +{
8
(
8
8 S- `z
8
B-GHh0
*L
8
0ī-pPK
Given desired goals, design a game (strategy sets and payoffs) in such a clever way that individual players, motivated solely by self-interest, end up achieving the designer's goals.(233R
8
C*Aj0286068pH
80h ?8 33___PPT10i.S+D=' c=
@B +
( $(
$
$ S}z `-
H
$0h ? 33___PPT10i.)+D=' =
@B +c
zrT
(
T
T S- `z
J
T
C"A
tips{'GH
T0h ? 33___PPT10i.0!;+D=' =
@B +
( 0(
0
0 SĚz `-
H
00h ? 33___PPT10i. C#+D=' =
@B +
X*(
X
X
B/GHaP
P
*
X
bApenny_guy \
/
X
0/
DAny questions? 233H
X0h ?X 33___PPT10i.!Н+D=' =
@B +
FxZ}l36 qؘ/SglNhL>Cm0h]UE#H5j4!
J)Mi*UEJ5Jpݽ;81mY~of{͛7k~ZMŕ{ɠk,J ȴi" qRV~{/9=n^{X $aSME#-H2fPh{v*;tXژV@>:;O܈|Tsq,\MWRD*PҁhiAJ>w}g#~2rw8T@`!X,ˀ=Ĺ-Q-P+*Ak
X6
@#l6fA`q\!l><ъ#@e{
tkGiߍt6PvMLF;CWٚGaﮞC{+澽p
ͪqj\[rWd8bK!i2-OO_I?b(T{Z9
1 $X9̊9Zf;-sWbܾ{Gl T.vZPܫ!/+uN+cl[CUk]+.@PCbōŒY+L+OY[e#08nd%%>zĊ\]t֟%mH4ZKHʑ8{_}36ظ[|S-;Glzd=uӗwM9=jRҴx*
*L/q
ӋޤPhh`IT ֽ3aypP8BA;pxN85LbCBXSkY%͖IBZBlx]eI,؆uS.ReJ8rH8rTGh鴱;6IêTim-6Ɍ*jv+։ *ϛ~nYɩJ Y,f0eGnj]Ye,< x6Kg2XC+W6LM.jxX=}8qq۹.\7G̳UiaڮO%
߰ȰrTUUHs\ǝuǵTK,֤
z{\QxMh%:xVZoq&%,UY忐/Ԉ#[s㨅u~.uDU2[KehiN㤐#TsWv9S[:-,>df
ȳJi~;q$jOAnM+eFRP*z4Ŭ.Jx\UHC1vA+mW?.YTeoiQA)|PzW;J:*1")ل:+sr-H7RcGHﲅޓ{("?IQx341-4~Ch'v3>#Zi{g)]QʲXn|&>1]QwOq|,z\".ȱc
eh=0[űc%vh=։OVqLH撷wNp-1"/Vjo\IӪ=mBmM[\qSBO ?zO >85=hzI|G~KTIMΊE4*GY̢NNH|,ȿD4'XRVz
ߧǫ[e]W????24hGNTe(XƊCIYJc}#g^eŪy]UPLCySQj@HMQZ;~0SS?1mrT3k=?E-l!X^p|7sB
x\͆K`S~*?
3(
83
Chart MSGraph.Chart.80*Microsoft Graph Chartb/0DArialSans}8 08z[ 00"DComic Sans MS}8 08z[ 00B
BSlide 5Slide 6Slide 7Slide 8Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16Fonts UsedDesign TemplateEmbedded OLE Servers
Slide Titles!5Gg _ ChristosChristos--@ !x--'-.
@n?" dd@ @@`` X_ 3
_b$$7{?No"$1VIG
<\Yob$6=>ݗE*Ou2$e)e튨hC&)Kz2$iC@@W[Ʋt0AAPf33333@g4DdDdPz[ 0ppp@<4!d!d| 0PW<4BdBd|< 0ʚ;ʚ;<4dddd||- 0h0___PPT10
N___PPT90(?
%,
d
Game Theory and Computer Networks:
a useful combination?
Christos Samaras, COMNET Group, DUTH8e;33%@
Game Theory:
general theory of rational behavior
What constitutes/characterizes a game:
group of players
(more than 1 decision-maker)
interaction
(interdependence: when a player acts, at least one other player is affected)
strategic
(a player takes this interdependence into account before deciding what action
to take)
rational
(a player chooses her best action)
VZ- --P--]--'-33'f
fPf]f&
consider...
some players that simultaneously transmit data
(a transport entity = a player)
they all share the same network channel
(low bandwidth)
each player chooses a strategy
(e.g. conservative/aggressive behavior)
every player is rational
Let the game begin !
(but... what are the rules of the game?)-3333
338 33w 33f
Visiting again our Internet Game, we have some answers:
WHO& ? a transport protocol (= an Internet player)
WHAT& ? transmit data
(aggressively, conservatively, etc.)
WHEN& ? no turns, no rounds&
(asynchronously)
HOW MUCH& ? gain? lose?
Utility Function
Z
3333 +33
:%33#F33S
Utility Function (= Payoff Function)
& specifies the PAYOFF to a player for every possible strategy combination that he and the others might pick. (Internet game: a player's payoff depends on his transmission rate and congestion/delay experienced by his packets)
The independent variable in such a function is the player s strategy.
So& what s the best strategy?
dp$33f
Solutions...
(to a game)
D!33
|Game Variations
Extensive Games
... players take turns; the possible actions a player can take on his turn depend on the previous actions taken by himself and the other players
Repeated Games
... a standard game that is played repeatedly
Signaling Games
... the players can signal each other and share some of their intended play, or private information
Bargaining Games
... includes states for making binding offers which the other player(s) can reject or accept
Games with Incomplete Information
... the players have to take actions without having full information on the different factors that influence their utility
$}P333333333.3333e3333^#3333|
Internet Equilibrium
Game: Internet congestion control
Some characteristics of the Internet Game :
repeated games
distributed environment
conditions and number of players change rapidly
Internet equilibrium is
not achieved by rational contemplation
but by interaction and adaptation
...in an environment where conditions (and player populations) change rapidly (and in which changes in strategies incur costs).
7Z.Z]-Z-ZZ7330f f0f336fH33
j
Nash equilibrium & is not (always) the solution to look for
Why?
efficiency, fairness are NOT GUARANTEED by Nash equilibrium
actually, a standard solution concept (like Nash equilibrium)
DOES NOT APPLY to an asynchronous and distributed
environment like the Internet
MORE SOPHISTICATED concepts of equilibria are
more appropriate for the context of the Internet
&H-m33+3
333fCfF
f[
A "perfect" design:
4
A (more or less) tangible goal:
Develop a reasonably faithful game-theoretic model of Internet congestion control for which (an approximation of) TCP/IP is a Nash equilibrium.
To be defined...
PAYOFF to each player (for instance: a player tries more but gains less)
what kind of PENALTY for selfish/greedy players?
WHO is responsible for penalizing a player? and HOW? (network: regulating role?)
desired GOALS:
1. for the network/Internet (better bandwidth utilization?
lower delays provision? fairness above all?)
2. for the players (better throughput/goodput? less packet losses/packet
retransmissions? smooth data transfer?)-V- -334f33R3333-33(33Vr
tSome facts about the Internet Congestion Game
not a "common knowledge" game
asynchronous, repeated games in a distributed environment
payoff function not known (nor the payoff functions of the opponents)
little a priori info (about other players/payoff function(s))
little or no knowledge of the underlying network topology
conditions change constantly
0 <H
!"#$%&'()*+,-./0123456789:;<=>?@ABCEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~Root EntrydO)`"@PicturescCurrent User68SummaryInformation(D
PowerPoint Document(DtDocumentSummaryInformation89.
@n?" dd@ @@`` X_ 3
_b$$7{?No"$1VIG
<\Yob$6=>ݗE*Ou2$e)e튨hC&)Kz2$iC@@W[Ʋt0AAPf33333@g4DdDdPz[ 0ppp@<4!d!d| 0X}<4BdBd|< 0ʚ;ʚ;<4dddd||- 0h0___PPT10
N___PPT90(?
%,
d
Game Theory and Computer Networks:
a useful combination?
Christos Samaras, ComNet Group, DUTH8e;33%,@
Game Theory:
general theory of rational behavior
What constitutes/characterizes a game:
group of players
(more than 1 decision-maker)
interaction
(interdependence: when a player acts, at least one other player is affected)
strategic
(a player takes this interdependence into account before deciding what action
to take)
rational
(a player chooses her best action)
VZ- --P--]--'-33'f
fPf]f&
consider...
some players that simultaneously transmit data
(a transport entity = a player)
they all share the same network channel
(low bandwidth)
each player chooses a strategy
(e.g. conservative/aggressive behavior)
every player is rational
Let the game begin !
(but... what are the rules of the game?)-3333
338 33w 33f
Visiting again our Internet Game, we have some answers:
WHO& ? a transport protocol (= an Internet player)
WHAT& ? transmit data
(aggressively, conservatively, etc.)
WHEN& ? no turns, no rounds&
(asynchronously)
HOW MUCH& ? gain? lose?
Utility Function
Z
3333 +33
:%33#F33S
Utility Function (= Payoff Function)
& specifies the PAYOFF to a player for every possible strategy combination that he and the others might pick. (Internet game: a player's payoff depends on his transmission rate and congestion/delay experienced by his packets)
The independent variable in such a function is the player s strategy.
So& what s the best strategy?
dp$33f
Solutions...
(to a game)
D!33
|Game Variations
Extensive Games
... players take turns; the possible actions a player can take on his turn depend on the previous actions taken by himself and the other players
Repeated Games
... a standard game that is played repeatedly
Signaling Games
... the players can signal each other and share some of their intended play, or private information
Bargaining Games
... includes states for making binding offers which the other player(s) can reject or accept
Games with Incomplete Information
... the players have to take actions without having full information on the different factors that influence their utility
$}P333333333.3333e3333^#3333|
Internet Equilibrium
Game: Internet congestion control
Some characteristics of the Internet Game :
repeated games
distributed environment
conditions and number of players change rapidly
Internet equilibrium is
not achieved by rational contemplation
but by interaction and adaptation
...in an environment where conditions (and player populations) change rapidly (and in which changes in strategies incur costs).
7Z.Z]-Z-ZZ7330f f0f336fH33
j
Nash equilibrium & is not (always) the solution to look for
Why?
efficiency, fairness are NOT GUARANTEED by Nash equilibrium
actually, a standard solution concept (like Nash equilibrium)
DOES NOT APPLY to an asynchronous and distributed
environment like the Internet
MORE SOPHISTICATED concepts of equilibria are
more appropriate for the context of the Internet
&H-m33+3
333fCfF
f[
A "perfect" design:
4
A (more or less) tangible goal:
Develop a reasonably faithful game-theoretic model of Internet congestion control for which (an approximation of) TCP/IP is a Nash equilibrium.
To be defined...
PAYOFF to each player (for instance: a player tries more but gains less)
what kind of PENALTY for selfish/greedy players?
WHO is responsible for penalizing a player? and HOW? (network: regulating role?)
desired GOALS:
1. for the network/Internet (better bandwidth utilization?
lower delays provision? fairness above all?)
2. for the players (better throughput/goodput? less packet losses/packet
retransmissions? smooth data transfer?)-V- -334f33R3333-33(33Vr
tSome facts about the Internet Congestion Game
not a "common knowledge" game
asynchronous, repeated games in a distributed environment
payoff function not known (nor the payoff functions of the opponents)
little a priori info (about other players/payoff function(s))
little or no knowledge of the underlying network topology
conditions change constantly
0 <H@<3333$33G
a
what next?
The aforementioned scheme calls for network rules and actions
(which will motivate the behavior of players)
...in this context we have selected TCP Vegas (throughput estimation mechanism)
remarks / ideas for proceeding:
network/Internet should not be vulnerable to greedy users
Nash equilibria are not necessarily achieved (as a result of learning)
nature of learning and convergence in the Internet
reasonable "learning" behavior-<--I--5--!- 0
p(
S/P `/
X
0X
0
X
0* X
0pXB
0DfԔ!
H
0h ? 33___PPT10i.S\J+D=' =
@B +r8?
3(
93
Chart MSGraph.Chart.80*Microsoft Graph Chartb/0DArialSansxW8 08z[ 00DComic Sans MSxW8 08z[ 00B
B
!"#$%&'()*+,-./012345:;<=>?@ABCDOh+'0
`hx
Slide 1 Christos Christos 82iMicrosoft PowerPointP@0bp@MV@P`"5Gg rx--@ !x--'-@ IZ>% -B( UUUU--r$7%T%P'L(I+F.D1B5A9@@,(%!!%(,"%(*,-.M.MALEKIILGODRAT=U9U5U1T.R+O(L'I%E%A%T--
@ IZ>%--@ IZJ8 -B( UUUU--8$9h=gAfDdGbI_K\LXMTLRLPKNJMHLGKEJCJAJ?K=L<M;N:P9R9T9h--
@ IZJ8--@ IZ)) -B( UUUU--X$*,(%!!%(,047:=?@@,.0235566655320.,--
@ IZ))---r%7%T%P'L(I+F.D1B5A9@@,(%!!%(,"%(*,-.M.MALEKIILGODRAT=U9U5U1T.R+O(L'I%E%A%T----H%"%T%X'\(_+b.d1f5g9h=gAfDdGbI_K\LXMTLRLPKNJMHLGKEJCJAJ?K=L<M;N:P9R9T9h----%MTM.----%
,047:=?@@@----(%,.0235566655320.,@--'@BComic Sans MS-. 33:2
td"Game Theory and Computer Networks:
."System-@BComic Sans MS-. 33'2
a useful combination?
.-@Arial-. 2
uChristos
.-@Arial-. 02
Samaras, COMNET Group, DUTH
.- 2
ComNet
.- _tChristosChristos-. 2
&Group, DUTH
.-՜.+,0
$On-screen Show n-stA
ArialComic Sans MSDefault DesignMicrosoft Graph ChartSlide 1Slide 2Slide 3Slide 4@<3333$33G
a
what next?
The aforementioned scheme calls for network rules and actions
(which will motivate the behavior of players)
...in this context we have selected TCP Vegas (throughput estimation mechanism)
remarks / ideas for proceeding:
network/Internet should not be vulnerable to greedy users
Nash equilibria are not necessarily achieved (as a result of learning)
nature of learning and convergence in the Internet
reasonable "learning" behavior-<--I--5--!- 0
w0(
l"
6"` `r
S@`
H
0h ? 33___PPT10i.7+D=' =
@B +rD ?
3(
93
Chart MSGraph.Chart.80*Microsoft Graph Chartb/0DArialSansxW8 08z[ 00DComic Sans MSxW8 08z[ 00B
B.
@n?" dd@ @@`` X_ 3
_b$$7{?No"$1VIG
<\Yob$6=>ݗE*Ou2$e)e튨hC&)Kz2$iC@@W[Ʋt0AAPf33333@g4DdDdPz[ 0ppp@<4!d!d| 0PW<4BdBd|< 0ʚ;ʚ;<4dddd||- 0h0___PPT10
N___PPT90(?
%,
d
Game Theory and Computer Networks:
a useful combination?
Christos Samaras, COMNET Group, DUTH8e;33%@
Game Theory:
general theory of rational behavior
What constitutes/characterizes a game:
group of players
(more than 1 decision-maker)
interaction
(interdependence: when a player acts, at least one other player is affected)
strategic
(a player takes this interdependence into account before deciding what action
to take)
rational
(a player chooses her best action)
VZ- --P--]--'-33'f
fPf]f&
consider...
some players that simultaneously transmit data
(a transport entity = a player)
they all share the same network channel
(low bandwidth)
each player chooses a strategy
(e.g. conservative/aggressive behavior)
every player is rational
Let the game begin !
(but... what are the rules of the game?)-3333
338 33w 33f
Visiting again our Internet Game, we have some answers:
WHO& ? a transport protocol (= an Internet player)
WHAT& ? transmit data
(aggressively, conservatively, etc.)
WHEN& ? no turns, no rounds&
(asynchronously)
HOW MUCH& ? gain? lose?
Utility Function
Z
3333 +33
:%33#F33S
Utility Function (= Payoff Function)
& specifies the PAYOFF to a player for every possible strategy combination that he and the others might pick. (Internet game: a player's payoff depends on his transmission rate and congestion/delay experienced by his packets)
The independent variable in such a function is the player s strategy.
So& what s the best strategy?
dp$33f
Solutions...
(to a game)
D!33
|Game Variations
Extensive Games
... players take turns; the possible actions a player can take on his turn depend on the previous actions taken by himself and the other players
Repeated Games
... a standard game that is played repeatedly
Signaling Games
... the players can signal each other and share some of their intended play, or private information
Bargaining Games
... includes states for making binding offers which the other player(s) can reject or accept
Games with Incomplete Information
... the players have to take actions without having full information on the different factors that influence their utility
$}P333333333.3333e3333^#3333|
Internet Equilibrium
Game: Internet congestion control
Some characteristics of the Internet Game :
repeated games
distributed environment
conditions and number of players change rapidly
Internet equilibrium is
not achieved by rational contemplation
but by interaction and adaptation
...in an environment where conditions (and player populations) change rapidly (and in which changes in strategies incur costs).
7Z.Z]-Z-ZZ7330f f0f336fH33
j
Nash equilibrium & is not (always) the solution to look for
Why?
efficiency, fairness are NOT GUARANTEED by Nash equilibrium
actually, a standard solution concept (like Nash equilibrium)
DOES NOT APPLY to an asynchronous and distributed
environment like the Internet
MORE SOPHISTICATED concepts of equilibria are
more appropriate for the context of the Internet
&H-m33+3
333fCfF
f[
A "perfect" design:
4
A (more or less) tangible goal:
Develop a reasonably faithful game-theoretic model of Internet congestion control for which (an approximation of) TCP/IP is a Nash equilibrium.
To be defined...
PAYOFF to each player (for instance: a player tries more but gains less)
what kind of PENALTY for selfish/greedy players?
WHO is responsible for penalizing a player? and HOW? (network: regulating role?)
desired GOALS:
1. for the network/Internet (better bandwidth utilization?
lower delays provision? fairness above all?)
2. for the players (better throughput/goodput? less packet losses/packet
retransmissions? smooth data transfer?)-V- -334f33R3333-33(33Vr
tSome facts about the Internet Congestion Game
not a "common knowledge" game
asynchronous, repeated games in a distributed environment
payoff function not known (nor the payoff functions of the opponents)
little a priori info (about other players/payoff function(s))
little or no knowledge of the underlying network topology
conditions change constantly
0 <H@<3333$33G
a
what next?
The aforementioned scheme calls for network rules and actions
(which will motivate the behavior of players)
...in this context we have selected TCP Vegas (throughput estimation mechanism)
remarks / ideas for proceeding:
network/Internet should not be vulnerable to greedy users
Nash equilibria are not necessarily achieved (as a result of learning)
nature of learning and convergence in the Internet
reasonable "learning" behavior-<--I--5--!- 0
r?
43(
:3
Chart MSGraph.Chart.80*Microsoft Graph Chartb/0DArialSans 8 08z[ 00DComic Sans MS 8 08z[ 00B
B.
@n?" dd@ @@`` X_ 3
_b$$7{?No"$1VIG
<\Yob$6=>ݗE*Ou2$e)e튨hC&)Kz2$iC@@W[Ʋt0AAPf33333@g4DdDdPz[ 0ppp@<4!d!d| 0` <4BdBd|< 0ʚ;ʚ;<4dddd||- 0h0___PPT10
N___PPT90(?
%,
d
Game Theory and Computer Networks:
a useful combination?
Christos Samaras, COMNET Group, DUTH8e;33%@
Game Theory:
general theory of rational behavior
What constitutes/characterizes a game:
group of players
(more than 1 decision-maker)
interaction
(interdependence: when a player acts, at least one other player is affected)
strategic
(a player takes this interdependence into account before deciding what action
to take)
rational
(a player chooses her best action)
VZ- --P--]--'-33'f
fPf]f&
consider...
some players that simultaneously transmit data
(a transport entity = a player)
they all share the same network channel
(low bandwidth)
each player chooses a strategy
(e.g. conservative/aggressive behavior)
every player is rational
Let the game begin !
(but... what are the rules of the game?)-3333
338 33w 33f
Visiting again our Internet Game, we have some answers:
WHO& ? a transport protocol (= an Internet player)
WHAT& ? transmit data
(aggressively, conservatively, etc.)
WHEN& ? no turns, no rounds&
(asynchronously)
HOW MUCH& ? gain? lose?
Utility Function
Z
3333 +33
:%33#F33S
Utility Function (= Payoff Function)
& specifies the PAYOFF to a player for every possible strategy combination that he and the others might pick. (Internet game: a player's payoff depends on his transmission rate and congestion/delay experienced by his packets)
The independent variable in such a function is the player s strategy.
So& what s the best strategy?
dp$33f
Solutions...
(to a game)
D!33
|Game Variations
Extensive Games
... players take turns; the possible actions a player can take on his turn depend on the previous actions taken by himself and the other players
Repeated Games
... a standard game that is played repeatedly
Signaling Games
... the players can signal each other and share some of their intended play, or private information
Bargaining Games
... includes states for making binding offers which the other player(s) can reject or accept
Games with Incomplete Information
... the players have to take actions without having full information on the different factors that influence their utility
$}P333333333.3333e3333^#3333|
Internet Equilibrium
Game: Internet congestion control
Some characteristics of the Internet Game :
repeated games
distributed environment
conditions and number of players change rapidly
Internet equilibrium is
not achieved by rational contemplation
but by interaction and adaptation
...in an environment where conditions (and player populations) change rapidly (and in which changes in strategies incur costs).
7Z.Z]-Z-ZZ7330f f0f336fH33
j
Nash equilibrium & is not (always) the solution to look for
Why?
efficiency, fairness are NOT GUARANTEED by Nash equilibrium
actually, a standard solution concept (like Nash equilibrium)
DOES NOT APPLY to an asynchronous and distributed
environment like the Internet
MORE SOPHISTICATED concepts of equilibria are
more appropriate for the context of the Internet
&H-m33+3
333fCfF
f[
A "perfect" design:
4
A (more or less) tangible goal:
Develop a reasonably faithful game-theoretic model of Internet congestion control for which (an approximation of) TCP/IP is a Nash equilibrium.
To be defined...
PAYOFF to each player (for instance: a player tries more but gains less)
what kind of PENALTY for selfish/greedy players?
WHO is responsible for penalizing a player? and HOW? (network: regulating role?)
desired GOALS:
1. for the network/Internet (better bandwidth utilization?
lower delays provision? fairness above all?)
2. for the players (better throughput/goodput? less packet losses/packet
retransmissions? smooth data transfer?)-V- -334f33R3333-33(33Vr
tSome facts about the Internet Congestion Game
not a "common knowledge" game
asynchronous, repeated games in a distributed environment
payoff function not known (nor the payoff functions of the opponents)
little a priori info (about other players/payoff function(s))
little or no knowledge of the underlying network topology
conditions change constantly
0 <H@<3333$33G
a
what next?
The aforementioned scheme calls for network rules and actions
(which will motivate the behavior of players)
...in this context we have selected TCP Vegas (throughput estimation mechanism)
remarks / ideas for proceeding:
network/Internet should not be vulnerable to greedy users
Nash equilibria are not necessarily achieved (as a result of learning)
nature of learning and convergence in the Internet
reasonable "learning" behavior-<--I--5--!- 0
L
(
L
L SI `
L
0I"`
Dominant Strategy: strategy Si strongly dominates all other strategies of player i if the payoff to Si is strictly greater than the payoff to any other strategy, regardless of which strategy is chosen by the other player(s). (the best strategy)X 233
G
>Fp
L
0xI"`O
t
Dominance Solvability: a strategy S1 is dominated by another strategy S2, if the latter does at least as well as S1 against every strategy of the other players, and against some it does strictly better. (try to find an undominated strategy: it s a good choice), 233
L
0I"`P
9
"Nash Equilibrium: a player selects the best strategy (which yields him the highest payoff possible) assuming what his opponents' strategy choice will be. A strategy combination (which comprises a strategy choice for each player) is a Nash equilibrium if each player s strategy is a best response against his opponents choices in that combination.,] 233Jw
dB
L
<DԔPdB
L
<DԔPdB
L
<DԔG
H
L0h ? 33___PPT10i. +D=' =
@B +r4h?
4t**