On the Correlation of Geographic and Network Proximity at Internet Edges and its Implications for Mobile Unicast and Multicast Routing

0.33 MB·English

by  Thomas C. Schmidt

On the Correlation of Geographi
and Network Proximity at Internet∗
Edges and its Impli
ations for Mobile Uni
ast and Multi
ast Routing.
†
Thomas C. S
hmidt Matthias Wählis
h Ying Zhang
t.s
hmidtieee.org waehlis
hieee.org zhang.yingweb.de
7 HAW Hamburg, Dep. Informatik, Berliner Tor 7, D(cid:21)20099 Hamburg, Germany
0
0
2

n
a Abstra
t 1. Introdu
tion
J

0 Signi(cid:28)
ant e(cid:27)ort has been invested re
ently to a
-
2
elerate handover operations in a next generation mo- Mobile environments, devi
es and appli
ations are
one of the major driving for
es for te
hnologi
al de-
bile Internet. Corresponding works for developing e(cid:30)-
] velopment today, while deployment is still dominated
I
ientmobilemulti
astmanagementareemergent. Both
N by non(cid:21)IP applian
es. However, the roadmaps of
on-
problems simultaneously expose routing
omplexity be-
s. tweensubsequentpointsofatta
hmentasa
hara
teris- vergedservi
esfor Next Generation Networks(NGNs)
[16℄ on the one hand, e(cid:30)
ient mobility management
c ti
parameter for handover performan
e in a

ess net-
[ within the next generation Internet [5℄ on the other,
works.
leadexpe
tationstotheInternetlayerastheprevalent
1 As
ontinuous mobility handovers ne
essarily o

ur
tie for mobile a

ess te
hnologies and servi
es.
v between a

ess routers lo
ated in geographi
vi
inity,
0 Seamless support for Voi
e over IP (VoIP) and
this paper investigates on the hypothesis that geograph-
3 related real(cid:21)time
ommuni
ation must be
onsidered
i
ally adja
ent edge networks attain a redu
ed network
1

riti
al for deployment su

ess into the mobile world.
1 distan
es as
ompared to arbitrary Internet nodes. We
Thereforesigni(cid:28)
ante(cid:27)ortis
ontinuouslytakeninthe
0 therefore evaluate and analyze edge distan
e distribu-
7 tionsinvariousregionsfor
lusteredIPrangeson their IETF to develop and improve proto
ols for seamless
0 mobility handovers, FMIPv6 [6℄ and HMIPv6 [14℄ be-
geographi
lo
ation su
h as a
ity. We use tra
eroute
/ ing the most prominent examples. The IP layer in-
s to
olle
t pa
ket forwarding path and round-trip-time
c trodu
ess
alablemulti
astassupplementaryfun
tion,
of ea
h intermediate node to s
an-wise derive an upper
:
v bound of the node distan
es. Results of di(cid:27)erent s
an- whi
h will be of parti
ular importan
e to multimedia
i group
onferen
es in mobile environments of limited
X ning origins are
ompared to obtain the best estima-

apa
ities. Seamless mobility extensions to IP(cid:21)layer
tion of network distan
e of ea
h pair. Our results are
r
a
ompared with
orresponding analysis of CAIDA Skit- multi
ast are likewise under preparation [11℄.
ter data, overall leading to fairly stable, reprodu
ible Mobile IPv6 inherits a strong topology dependen
e
edgedistan
edistributions. Asa(cid:28)rst
on
lusiononex- through its binding update pro
edures with the Home
pe
ted impa
t on handover performan
e measures, our Agent (HA) and the Correspondent Node (CN). Han-
results indi
ate a general optimum for handover anti
- dover a

eleration s
hemes attempt to over
ome this
ipation time in 802.11 networks of 25 ms. obsta
le by relo
ating immediate transfer negotiations
to the vi
inity of the mobile node, i.e., to a

ess net-
works at the Internet edges. In previous analysis [9℄
it
ould be shown that the a
tually attained handover
Keywords: Internet S
anning, Measurement, Edge
performan
e largely depends on the relative network
Distan
e Distribution, Mobile Handover Performan
e,
topology of a

ess
omponents, when measured in an
Mobile Multi
ast
appropriate delay metri
su
h as round trip time. A
-
∗
This work is supported by the German Bundesministerium
essrouterdistan
e
anbe
onsideredasthe
hara
ter-
für Bildung und Fors
hung within the Proje
t Movie
ast. The
isti

omplexity parameter in fast or hierar
hi
al mo-
aut†horsthankCAIDAformakingSkitter datasetsavailable.
bile IPv6.
Theauthorisalsowithlink(cid:21)lab,HönowerStr. 35,D(cid:21)10318
Berlin,Germany. Similar observations hold for mobile multi
ast.
Source: 141.22.64.9
While multi
ast listeners may rely on handover oper-
ations derived from uni
ast proto
ols, mobile multi-

astsour
esneedto reshapedistributiontrees. Due to HHoopp 11:: 114411..2222..6644..11
theself(cid:21)similarnatureofshortestpathtrees,alteration
requirements of multi
ast forwarding states are mini- HHoopp 22:: 114411..2222..44..112211
mallybound by a
hara
teristi
fun
tion of sour
e dis-
pla
ement. As shownin[10℄ the routing
ostofmobile HHoopp 33:: 118888..11..4477..5577
multi
ast sour
e management in SPT-based proto
ols
is dire
tly dominated by the topologi
al hop distan
e
HHoopp 44:: 118888..11..1188..110099
attained between designated routers at the previous
and next point of atta
hment. HHoopp 55:: 118888..11..1188..115544
Inthisworkweempiri
allyanalyzetheregionaledge
distan
edistributionsofexemplaryareasinthe
urrent HHoopp 66:: 118888..11..8800..4466 HHoopp 66::118811..11..8800..6666
Internet. We perform sele
tive s
ans and evaluate the
TTrraannssiitt ppooiinntt
datapooloftheCAIDASkitterproje
tandderivemo- HHoopp 77:: 8800..8811..119922..220088
bility performan
e spe
tra thereof. The paper is orga-
HHoopp 88:: 119944..114400..111111..222277 HHoopp 88:: 119944..114400..111133..9955
nized asfollows. We introdu
eourevaluationmethod-
ologies,dis
uss problems, limitations and related work
HHoopp 99::221122..6633..4466..8877
in the following se
tion. Measurement results are pre-
sented and dis
ussed in se
tion 3. Se
tion 4 brie(cid:29)y
Host 1: 62.8.134.1
derives
onsequen
es for
hara
teristi
handover per-
Network 1 Host 2: 62.8.129.1
forman
e measures of our data. Finally, se
tion 5 is Network 2
dedi
ated to
on
lusions and an outlook.
Figure 1. Transit Point Discovery withtracer-
oute
2. Methodology and Related Work
2.1. Edge Distance Estimates
assumption) and from the perspe
tive of the sour
e.
The obje
tive of this work is to inquire on the dis-
S
anexperiments thereforearerepeated from multiple
tribution of network distan
es between pairs of Inter-
origins,eventuallydis
overing
losertransitpointsand
net edge routers lo
ated within geographi
vi
inity.
shorterneighboringpaths asdisplayed in (cid:28)gure 2. Re-
A

ording to the mobility investigations introdu
ed
sults are then derived as minimal upper bound using
above, network distan
e measures require metri
s of
the path via the
losest transit point.
delayandrouting
omplexity,e.g.,roundtriptimeand
hop
ount. S
anning
an thus be performed by tra
er- Geographi
lo
ationsofIPaddressrangesneedpro-
oute and will pro
eed as follows. visioning as external input to the s
anning. To
hoose
Clusters of IP ranges from geographi
regions su
h forareliablesour
eofgeographi
informationweevalu-
as
itiesarepre-sele
tedinordertoa

ountforlo
ality. atedeightdi(cid:27)erentmappingresour
esina(cid:28)rststepby
Alifehostisdeterminedfromea
hrange,su
hthatfor sele
ting a set of 30 distributed, geographi
allyknown
ea
h pair of hosts the network distan
e of their
orre- IP ranges. The
ommer
ial produ
t GeoIP [7℄ thereby
sponding a

ess routers
an be evaluated as a sample was the only resour
e to admit negligible errors. In a
probe. Theappropriatemethodtodeterminearouting se
ond, automated testing we
ompared data of larger
pathbetweentwoInternethostsfromaremotelo
ation samples with whois queries and found a
oin
iden
e
isgivenbyloosesour
erouting. Sin
esour
eroutingis rate of about 80 %. This result we
onsidered reason-
mainly unsupported throughout the Internet, we pro- able, as whois data
ommonly provide administrative

eed by auxiliary means as shown in (cid:28)gure 1: Both addressespossiblydistin
tofphysi
alrouterlo
ations.
routingpathfromtheorigintothehostsunder
onsid- It should be mentioned that measurements have
erationare evaluated via tra
eroute [3℄ and
ompared. been undertaken forIPv4 only, even though the initial
Thelast
ommonhopthenisdeterminedasapotential motivation was derived from mobile IPv6 handovers.
transitpoint,givingrisetoanupperbound ofthenet- Due to our limited a

essibility of IPv6
ommuni
a-
workpathbetweenthe targetnodesviathedis
overed tion this pro
edureis justi(cid:28)ed by the expe
tation that
transit point. Su
h Evaluation is done under the as- quantitatively dominating dual sta
k a

ess networks
sumption of symmetri
routing (
f. se
tion 2.2 on this will attain identi
al topologi
al properties.
Source 1
The Mer
ator proje
t [1℄ uses hop-limited probes in
Source 2
the same manner as tra
eroute, to infer an Internet
TTrraannssiitt ppooiinntt 11 map at the route-level. A highly distributed s
anning
TTrraannssiitt ppooiinntt 12
approa
his taken within the DIMES proje
t [12℄ with
the aim of in
reased a

ura
y and
omprehensiveness.
Aside from many others, Jani
and Van Mieghem [4℄
re
ently performed tra
eroute s
ans to investigate the
node degree distribution in the Internet and report
about
ompli
a
ies similar to our observations.
Sin
e IP addresses are lo
ation-independent, there
HHoosstt 11:: 221177..8899..4477..111133 has been mu
h work on the problem of
orrelating IP
NNeettwwoorrkk 11 HHoosstt 22:: 221177..8833..118855..110099 addresses to geographi
lo
ations. One of the latest
NNeettwwoorrkk 22
studieswasdonebySubramanianetal. [15℄. Theysug-
gest that 90 % of nodes within 5 ms RTT are lo
ated
Figure 2. Approximation by Multisource
within a radius of 50 km and 90 % of nodes within
Scanning
10 ms RTT are lo
ated within a radius of 300 km.
A CAIDA group [2℄ has studied network
onne
tivity
in the Asia-Pa
i(cid:28)
region, mainly fo
using on network
2.2. Scanning Problems with traceroute laten
y and performan
e,
ountry peering and third
partytransit. Theyuseskitter to measuretheforward
IP path and round-trip-time to about 2,000 destina-
Thewellknowntra
erouteutility[3℄sendsoutTTL
tions in the Asia-Pa
i(cid:28)
region from di(cid:27)erent sour
es
restri
tedUDPprobepa
ketstoinquireonallforward-
and
on
ludethatgeographi
proximityre(cid:29)e
tsonlyin
ing nodes along a path to a given destination. It thus
round trip times, while hop
ount is not a representa-
leadstoa
ompletetopologi
alpathve
toroftheexam-
tive metri
therefor.
ined route. However, a fair number of nodes through-
out the Internet does not reply
orre
tly to standard
tra
eroute queries for several reasons. At (cid:28)rst, UDP 3. Regional Edge Distan
e Distributions
forwarding may be blo
ked at some router, in whi
h

ase tra
eroute
an be swit
hed to use ICMP probes.
Inthisse
tionwepresenttheresultsforthe
itiesof
Ifthelatterareblo
ked,aswell,probingissu

essfully
San Fran
is
o, USA, Berlin and Hamburg, Germany,
inhibited. Atse
ond,ICMPrepliesaresometimessup-
and Shanghai, China, whi
h were exemplarily sele
ted
pressed or manipulated by ISPs, whi
h only interferes
as geographi
target regions. S
anning has been per-
with our measurements if this happens at or beyond
formedfromSeptember to De
ember 2006,originating
a transit node. The o

urren
e of routing loops be-
from the lo
ations of Berlin, Hamburg and Shanghai,
yond the transit point will likewise lead to dis
arding
at the 67th IETF meeting in San Diego and with the
the event. The most frequent reason for obtaining in- 1
help of various publi
tra
eroute fa
ilities. Sin
e the
validdataresultsfromasymmetri
routes,dis
overable
numberofavailableIPrangesvaryfromShanghai(763)
by de
reasing
umulative round trip times. In a gen-
up to San Fran
is
o (8476), subsets of equal sizes are
eral attempt Paxson [8℄ analyzed 40,000 end(cid:21)to(cid:21)end
sele
tedrandomlyforea
h
ity. Statisti
al
onvergen
e
paths and identi(cid:28)ed half of them as asymmetri
. As
withrespe
tto samplesize, but alsofordi(cid:27)erentdates
ourexperimental
on
ern
on
entratesonedgetopolo-
and day times were
ompared, and a fair stability of
gies, measurements remain una(cid:27)e
ted by asymmetries
the distributions
ould be observed for sample subsets
within the
ore, leading to signi(cid:28)
antly lowereventre-
of 500 IP ranges.
je
tion rates. Altogether the su

ess ratio in our mea-
We
ompare our results with distributions derived
surements varies between 45 % and 65 %.
fromCAIDAdatare
ordedinO
tober2006. Host
lus-
2.3. Related Work ters for sele
ted
ities are taken from the CAIDA des-
tination list a

ording to the GeoIP database. Tra
e
paths are minimized with respe
t to all available 18
Internettopologyhasbeenstudiesforovertenyears.
monitor points, whi
h are lo
ated more densely at the
These studies fo
us on
hara
terizing and delineating
US West Cost and sparsely in Europe and Asia.
Internet topology and performan
e. CAIDA [13℄ is
a prominent group of pioneers, who re
ord and mea- 1Publi
tra
erouteservi
esdidnotprovideICMPprobingand
sure Internet data
ontinuously for almost ten years. resultantly providedonlylittlesu

essrates.
Berlin, Germany <x>=9,42273 x=2,99129 Berlin, Germany <x>=12,02614 x=5,67525
0,15 Hamburg, Germany <x>=8,00368 x=2,89005 0,15 Hamburg, Germany <x>=10,94393 x=4,98594
San Francisco, USA <x>=15,55914 x=5,80903 San Francisco, USA <x>=9,49885 x=5,33369
ncies SRhaanndgohma i , C h i n a <<xx>>==1195,,8267888577 xx==75,,7439897265 ncies SRhaanndgohma i , C h i n a <<xx>>==1142,,3542582721 xx==45,,7602908151
e e
qu0,10 qu0,10
e e
Fr Fr
e e
v v
ati ati
el el
R R
0,05 0,05
0,00 0,00
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Distance [Hops] Distance [Hops]
(a) S
an Data (b) CAIDAData
Figure 3. Hop Count Distributions at Network Edges in Four Cities. < x > and σx represent mean
andstandarddeviationofthecorrespondingdistributions.
0,05 0,05
Berlin, Germany <x>=86,29672 x=185,46638 Berlin, Germany <x>=94,14779 x=130,77141
Hamburg, Germany <x>=79,75878 x=175,09762 Hamburg, Germany <x>=73,43109 x=116,76854
0,04 0,04
San Francisco, USA <x>=73,1862 x=68,57574 San Francisco, USA <x>=21,08282 x=43,99535
es Shanghai, China <x>=388,07119 x=341,45648 es Shanghai, China <x>=257,61768 x=252,64217
enci0,03 Random Location <x>=277,49569 x=242,80439 enci0,03 Random <x>=150,74125 x=265,74948
u u
q q
e e
Fr Fr
e e
ativ0,02 ativ0,02
el el
R R
0,01 0,01
0,00 0,00
0 50 100 150 200 250 300 0 50 100 150 200 250 300
RTT [ms] RTT [ms]
(a) S
anData (b) CAIDAData
Figure4.RoundTripTimeDistributionsatNetworkEdges
Resulting distributions for hop
ounts and round San Fran
is
o, e.g., in San Jose, whereas signi(cid:28)
ant
trip times aredisplayed in (cid:28)gures3 and 4respe
tively. parts of our distribution was built from s
ans at San
Anadditional
urvederivedforrandomlylo
atednodes Diegoorigin. Fromthis argumentit maybe
on
luded
is added to distinguish lo
ality
orrelations. thatourdatabetterapproximateedgedistributions of
the European
ities, while CAIDA values re(cid:29)e
t those
Clearly our hop
ount results vary signi(cid:28)
antly, of San Fran
is
o in a better quality.
whileCAIDAskitterdataevaluatetofairlysimilardis-
tributions. For Berlin and Hamburg we measure
lear Hop
ounts in Shanghai admit mu
h wider distri-
peaksataround8hopsandunderestimatethosevalues butions, even though data are mainly re
orded from
of CAIDA, while we slightly overestimate hop
ounts a sour
e in Shanghai. S
ans originating from Europe
for San Fran
is
o. These di(cid:27)eren
es may be explained even show a geographi
anti(cid:21)
orrelation when
om-
by monitor point positions. While we had dire
t a
- pared to the random sample. These results indi
ate

ess to several networks lo
ated in Berlin and Ham- thatregionalroutingtopologiesarenotdenselymeshed
burg, skitter data are not available from any monitor in the Chinese
ity, su
h thatgeographi
allyneighbor-
point
loseto thetwo
itiesorevenin Germany. Some ing a

ess networks are mainly
onne
ted via a gen-
CAIDA monitor points are lo
ated in
lose vi
inity of eral degree of indire
tion. On the
ontrary, Berlin
0,25 In total the results seem to indi
ate that inter(cid:21)edge
Origin Hamburg, Germany routing within a geographi
region is frequently per-
Origin San Diego, USA
0,20 formed via lo
al transits and peering, whi
h produ
e
s network proximity in 'the neighborhood', but remain
e
nci invisible for a distant monitor.
e0,15
u
q
e
Fr
e
ativ0,10 4. Appli
ations to Handover Perfor-
el
R
man
e
0,05
The results obtained so far may serve as an empiri-
0,00

al fundament for realisti
handover performan
e esti-
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700
RTT [ms] matesofthenetwork. Amobilenode movingfromone
a

ess network to another in geographi
al neighbor-
Figure 5. RTT Distributions Evaluated from
hood does imposes tra(cid:30)
redire
tion, minimally from
DifferentOriginsforSanFranciscoNodes
its previous to its new atta
hment. These operations

ause delay and routing
osts, whi
h for the
ase of
FMIPv6 [6℄ are given by the uni
ast path from previ-
ous to next a

ess router, and higher, otherwise.
and Hamburg results expose a pronoun
ed lo
ation(cid:21)
Basedontheresultsderivedin[9℄wenow
animme-
to(cid:21)network hop distan
e
orrelation, whi
h somewhat
diately
al
ulate expe
ted values of
hara
teristi
han-

ontradi
ts the
on
lusions drawn in [2℄. Caida mon-
dovermeasures. Forpa
ketssentata
onstantbitrate
itors seem to be too sparsely distributed to
at
h a
of one per 10 ms, the
onditional expe
tation of pa
k-

lear, distinguishable view on hop
ount laws at Inter-
ets lost or bu(cid:27)ered for given inter(cid:21)a

ess(cid:21)router delay
net edges.
was derived for predi
tive and rea
tive handover pro-
Round trip time distributions exhibit similar be-

edures (
f. (cid:28)gure 6 of [9℄). Combining these previous
havior. Pronoun
ed peaks at
lose distan
es
an be
results with those shown in (cid:28)gure 4, we arrive at ex-
observed for the areas of Berlin and Hamburg, when
pe
ted periods forpa
ketlossasfun
tions of handover
monitored from the
lose vi
inity. The e(cid:27)e
t of s
an-
anti
ipation times. Results for the di(cid:27)erent regions as
ning sour
e positioning on RTT results is shown for
presented in (cid:28)gure 6 jointly show a pronoun
ed uni-
San Fran
is
o data in (cid:28)gure 5. RTT
hara
teristi
s,
form minimum at handover anti
ipation of 25 ms for
though, appear heavier tailed than hop
ounts, whi
h
the
ities ofSan Fran
is
o,Hamburgand Berlin, while
supposedly is due to sporadi
slow transition links.
signi(cid:28)
ant optimal values remain absent for Shanghai
Tardy transitions are of lesser e(cid:27)e
t in the San Fran-
and random data. These results re(cid:29)e
t the degree of

is
o region, for whi
h again CAIDA measurements
lo
ality in regional delay distributions.
segregate a distribution of higher signi(cid:28)
an
e.
Similar
on
lusions
anbedrawnfor
ostsneededto
In
ontrast, Shanghai data attain merely indi(cid:27)er-
reshape shortest path multi
ast distribution trees un-
entRTTdistributions,whi
hareevenlesspronoun
ed
der sour
e mobility. In previous simulation studies we
than random samples. Non-negligible weights are sit-
derived
onditionalexpe
tationsformulti
astforward-
uated beyond the displayed interval, as
an be read
ing state persisten
e at a given hop distan
e between
from mean and standard deviation values. This may
previous and next designated router (
f. (cid:28)gure 3 of
be explained from a wide variety of slow transit links
2 [10℄). Resolving
onditioningwiththehelpofabovere-
present in the Chinese
ore networks. CAIDA skit-
sults gives rise to an expe
ted ratioof multi
ast states
ter data seem to qualitatively re(cid:29)e
t these RTT law
persistentunder mobility fortheregionsunder
onsid-
diversities. San Fran
is
o values are very pronoun
ed,
eration. A mobile multi
ast sour
e feeding a region-
whereas Hamburg and Berlin data show an interme-
ally distributed group in San Fran
is
o, Hamburg or
diate
hara
teristi
. It should be noted, though, that
Berlinwillhavetoexpe
tonhandovertheinvalidation
thereverseoftheproximityobservationsfrom[15℄does
of about 5% of forwardingstates at multi
ast routers,
not seem to hold: RTT distributions admit wide tails,
while10%oftheroutersina
omparablehandoversit-
when
eevenin
loserouterdistan
esenhan
edmutual
uation in Shanghai need to establish new states. Cor-
delays may be expe
ted.
respondingly, from 75 % down to 60 % of multi
ast
2Anotherpossibleexplanation
ouldlieinaredu
eda

ura
y states survive a sour
e handover in widely distributed
ofMaxMindGeoIP datafortheChineseregion. groups.
130 COM 2000, volume 3, pages 1371(cid:21)1380, Pis
ataway,
NJ, USA, Mar
h 2000. IEEE Press.
120
[2℄ B.Hu(cid:27)aker,M.Fomenkov,D.Moore,E.Nemeth,and
110 k
la(cid:27)y. Measurements of the Internet topology in
ms]100 the Asia-Pa
i(cid:28)
Region. In INET 2000 Pro
eedings,
s [ Yokohama, Japan, June 2000. InternetSo
iety.
s
o
L 90 [3℄ V. Ja
obson. Tra
eroute tool.
et
ck ftp://ftp.ee.lbl.gov/tra
eroute.tar.gz, 1989.
Pa 80 [4℄ M. Jani
and P.VanMieghem. Onproperties ofmul-
70 Berlin, Germany ti
ast routing trees. Int. J. Commun. Syst.,19(1):95(cid:21)
Hamburg, Germany
San Francisco, USA 114, 2006.
60 Shanghai, China [5℄ D. B. Johnson, C. Perkins, and J. Arkko. Mobility
Random
50 Support in IPv6. RFC3775, IETF, June 2004.
0 25 50 75 100 [6℄ R. Koodli. Fast Handovers for Mobile IPv6. RFC
Anticipation Time [ms]
4068, IETF, July 2005.
Figure 6. Expected Packet Loss in Predic- [7℄ MaxMind LLC. GeoIP. http://www.maxmind.
om,
tive Handoversas a Functionof Anticipation 2006.
[8℄ V. Paxson. End-to-End Routing Behavior in the In-
Time(0=ReactiveHandover)
ternet. IEEE/ACMTrans.Netw.,5(5):601(cid:21)615, 1997.
[9℄ T. C. S
hmidt and M. Wählis
h. Predi
tive ver-
sus Rea
tive (cid:21) Analysis of Handover Performan
e
and Its Impli
ations on IPv6 and Multi
ast Mobil-
5. Con
lusions and Outlook ity. Tele
ommuni
ation Systems, 30(1(cid:21)3):123(cid:21)142,
November2005.
[10℄ T. C. S
hmidtand M. Wählis
h. Morphing Distribu-
Quality of servi
e
hara
teristi
sof the Internet are
tionTrees (cid:21) OntheEvolutionof Multi
astStatesun-
likelytodriveordelaythedeploymentofall(cid:21)IPmobile
derMobilityandanAdaptiveRoutingS
hemeforMo-
devi
es. Inspired by mobile IP handover performan
e
bile SSMSour
es. Tele
ommuni
ation Systems, O
to-
measures we analyzed routing distan
es in geograph-
ber 2006. in print,appeared online.
i
ally bound
lusters of the Internet, in whi
h users [11℄ T. C. S
hmidt and M. Wählis
h. Multi
ast Mobility
areexpe
ted to movearoundfreely while
ontinuously in MIPv6: Problem Statement. IRTF Internet Draft
'talking'IPin thenearfuture. Tra
erouteprobeshave (cid:21) work in progress 01, MobOpts, O
tober 2006.
been used to derive hop
ount and delay distributions [12℄ Y. Shavitt and E. Shir. DIMES: Let the Inter-
at Internet edges in San Fran
is
o, Berlin, Hamburg net Measure Itself. Te
hni
al Report
s.NI/0506099,
andShanghai. Comparisonhasbeen drawnto CAIDA arXiv.org, June 2005.
[13℄ Skitter Proje
t at CAIDA.
measurements. Our results seem to indi
ate a
lear
http://www.
aida.org/tools/measurement/skitter.
signatures of lo
ality in both distan
e metri
s, whi
h
[14℄ H. Soliman, C. Castellu

ia, K. Malki, and L. Bel-

annot be segregated from CAIDA measurements due
lier. Hierar
hi
al Mobile IPv6 mobility management
to sparsely s
attered monitor points. The appli
ation
(HMIPv6). RFC 4140, IETF, August2005.
of these results to
al
ulating pa
ket loss after mobil- [15℄ L. Subramanian,V.Padmanabhan,andR.Katz. Ge-
ity handovers indi
ates that
hara
teristi
proximity ographi
Properties of Internet Routing. In Pro
eed-
measures in the Internet may give rise to fairly stable ings of the 2002 USENIX Annual Te
hni
al Confer-
anti
ipation timers. en
e, pages 243(cid:21)259, Berkeley, CA, USA, June 2002.
In future work we will re(cid:28)ne our measures by tar- USENIXAsso
iation.
[16℄ Tele
ommuni
ations and Internet
onverged Servi
es
geting additional geographi
regions and adding Plan-
and Proto
ols for Advan
ed Networking (TISPAN);
etLab nodes to our monitors. We intend to in
lude
NGN Release 1; Release de(cid:28)nition. Te
hni
al Report
DIMES data into our
omparison,
ontinue to derive
180 001 V1.1.1, ETSI, Mar
h 2006.
QoS distributions and
hara
teristi
values from re-
gional delay distributions to put expe
ted handover

hara
teristi
s of a future mobile Internet on (cid:28)rmer
grounds.
Referen
es
[1℄ R. Govindan and H. Tangmunarunkit. Heuristi
s for
internet map dis
overy. In Pro
eedings IEEE INFO-