Measurements & Analysis · ~7,055 km Destination login.cs.duke.edu Duke Univ., 308 Research Dr.,...
Transcript of Measurements & Analysis · ~7,055 km Destination login.cs.duke.edu Duke Univ., 308 Research Dr.,...
Measurements&Analysis
QuickdiveintonetworklatencyProf.AnjaFeldmann
BalakrishnanChandrasekaran
May22,2017
§ Round-triptime(RTT)betweentwohosts.
2
Latency
OnLatency…
3
StuartCheshire,“It’stheLatency,Stupid”,http://www.stuartcheshire.org/rants/latency.html
Latencymatters…
4
http://geekandpoke.typepad.com/geekandpoke/2012/12/simply-explained-latency.html
ToyExperimentSource
tst1.inet.tu-berlin.de
Destinationlogin.cs.duke.edu
5
ToyExperimentSource
tst1.inet.tu-berlin.deTUB,Marchstraße 23,Berlin,DE
Destinationlogin.cs.duke.edu
DukeUniv.,308ResearchDr.,Durham,NC,USA
6
7
ToyExperiment
ToyExperiment
Sourcetst1.inet.tu-berlin.de
TUB,Marchstraße 23,Berlin,DE
~7,055km
Destinationlogin.cs.duke.edu
DukeUniv.,308ResearchDr.,Durham,NC,USA
8
ToyExperiment
Sourcetst1.inet.tu-berlin.de
TUB,Marchstraße 23,Berlin,DE
~7,055km;expectedRTT?~70.55ms
Destinationlogin.cs.duke.edu
DukeUniv.,308ResearchDr.,Durham,NC,USA
9
ToyExperimentmeasurement17@tst1:~$ ping -c 3 quicksort.cs.duke.eduPING quicksort.cs.duke.edu (152.3.140.29) 56(84) bytes of data.64 bytes from quicksort.cs.duke.edu (152.3.140.29): icmp_seq=1 ttl=48 time=114 ms64 bytes from quicksort.cs.duke.edu (152.3.140.29): icmp_seq=2 ttl=48 time=113 ms64 bytes from quicksort.cs.duke.edu (152.3.140.29): icmp_seq=3 ttl=48 time=113 ms--- login.cs.duke.edu ping statistics ---3 packets transmitted, 3 received, 0% packet loss, time 2003msrtt min/avg/max/mdev = 113.666/113.869/114.265/0.280 ms
10
ToyExperiment
Sourcetst1.inet.tu-berlin.de
TUB,Marchstraße 23,Berlin,DE
~7,055km;expectedRTT?~70.55msObservedRTT?~113.73ms(~1.61-times;~61%more)
Destinationlogin.cs.duke.edu
DukeUniv.,308ResearchDr.,Durham,NC,USA
11
Research
WhyistheInternetsoslow?!I.N.Bozkurtetal.,InProceedingsofthePassiveandActiveMeasurementsConference(PAM),2017.
12
InternetattheSpeedofLight
13
http://cspeed.net
FetchingHTMLofbasepages
Top500Websitesfromeachof138 countrieslistedbyAlexa§
22,800 URLsfetchedfrom102 PlanetLab° locationsusingcURL
cLatency— timeforlighttotravelround-tripalongtheshortestpathbetweentwoend-points
14
§ Top500sitesineachcountryorterritory,Alexa.http://goo.gl/R8HuN6°PlanetLab,https://www.planet-lab.org
PlanetLab|Inflationinfetchtime
0
0.2
0.4
0.6
0.8
1
1 10 100 1000
CDF
over
trac
es
Inflation over c-latency
15
36.5-times!
Ttotal =TDNS +Thandshake +Trequest +Tserverproc +Tresponse +Ttransfer
Ifnetwork’sRTTisreducedbyafactorofx …
everythingontherighthandside(exceptTserverproc)reducesbyafactorofx (toanapproximation)
16
Inflationbreakdown
PlanetLab|Inflationbreakdown
�
���
���
���
���
�
� �� ��� ����
�����������������
�������� ���� ���������
��������������� �������
��� �������������������������
��� ������������� ����
17
~3-times!
End-user-basedMeasurements
18
Coursera|CloudNetworking,https://www.coursera.org/learn/cloud-networking#
MOOCMeasurements
Studentsenrolledinamassiveopenonlinecourse(MOOC)offeredviaCoursera
50Webpagesperstudent
613 uniqueIPaddresses607 unique/24 blocks
Groundtruthdatawasprovidedbystudents
19
MOOCStudents
20
MOOC|Inflationbreakdown
0
0.2
0.4
0.6
0.8
1
1 10 100 1000
CDF
over
HTT
P tr
aces
Inflation over c-latency
Min. pingDNS
TCP handshakeRequest-Response
TCP transferTotal time
21
~4-times!
RIPEAtlas
Atlasplatformprobes§ — smallnetworkdevicesthataretypicallydeployedinend-usernetworks
PingsoverIPv4 (IPv6)betweenAtlasprobes288,425 (63,884)uniquesource-destinationpairs
85% (78%)areinter-AS
22
§RIPEAtlas.https://atlas.ripe.net
RIPEAtlas|Inflationinpings
�
���
���
���
���
�
� �� ���
���
�������� �� ���� ���� ���� ���������
����� ������������� ������������� ������������� ��������
23
RIPEAtlas|InflationvsDist.
�
��
���
� ���� ����� ����� ����� ����� ����� �����
������������������������
�������� ��� ���
�� ��������
24
MeasurementsfromaCDN
25
https://www.akamai.com/us/en/about/
PassiveMeasurements
Round-triptimebetweenCDNserversandendusers,measuredbytheservers
TimeelapsedbetweenTCP-SYN-ACK andTCP-ACK
26
CDN|Inflationinpings
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
1 10 100 1000
CD
F ov
er s
erve
r-cl
ient
pai
rs
Inflation over c-latency
Min RTTAvg RTTMax RTT
27
Research
AServer-to-ServerViewoftheInternetB.Chandrasekaranetal.,InProceedingsoftheACMConferenceonemergingNetworkingExperimentsandTechnologies(CoNEXT),2015.
28
Server-to-serverMeasurements
Measurements(traceroutes)overIPv4 andIPv6
Conductedbetween~600 servers(full-mesh)overaperiodof16months
29
CDNServers|Inflationinpings
00.10.20.30.40.50.60.70.80.91
1 10
ECDF
RTT/cRTT
IPv4IPv6
30
CDNServers|Measurements
Conducttraceroutesbetweenserverpairsfor16monthsoverIPv4andIPv6
WhatdotheRTTsbetweentheservers(endpoints)looklikewhenplottedasafunctionoftime?
31
CDNServers|LatencyTimeline
0
50
100
150
200
250
300
350
400
Jan Feb Mar AprApr May Jun Jul
RTT(in
ms)
IPv4IPv6
32
Asix-monthtimelineofround-triptimesbetweenserversinHongKong,CN andTokyo,JP
CDNServers|Latencypattern
50
100
150
200
250
300
350
03/26 03/27 03/28 03/29 03/30 03/31 04/01 04/02
RTT(in
ms)
IPv4IPv6
Night Day
33
DailyoscillationsinRTTbetweentheservers.
Usingmeasurementstounderstand…
Whatfactorscontributetolatency?Whydoeslatencychangedrasticallybetweenapairofendpoints?Whatdothedailyoscillationsinlatencyimply?Howoftendoeslatencychangedrastically?
34
Onlythetipoftheiceberg
Weworkextensivelyonawidevarietyofresearchtopicsspanningnetworkmeasurements,networksecurity,andsoftware-definednetworking.http://www.inet.tu-berlin.de/menue/research/publications/by_author/anja_publ/
35
Questions?
“Thereisnosuchthingasadumbquestion.”— CarlSagan
Ifyouarefeelingshy,emailyourquestions!{anja,balac}@inet.tu-berlin.de(withsubject[IM17]Q:…)
36