Roma Tre University
Is It Really Worth to Peer at IXPs?
Marco Di Bartolomeo Giuseppe Di Battista Roberto di Lallo Claudio Squarcella
Nautilius Mediterranean Exchange Point (NaMeX)
ISCC 2015
UNIVERSITÀ DEGLI STUDI
ROMATRE
Introduction to IXPsInternet eXchange Points (IXPs) are infrastructures used by Internet Service Providers (ISPs) to exchange traffic between their ASes
UNIVERSITÀ DEGLI STUDI
ROMATRE
Known Benefits
UNIVERSITÀ DEGLI STUDI
ROMATRE
Known Benefits• Lower distances
and locality of traffic
UNIVERSITÀ DEGLI STUDI
ROMATRE
Known Benefits• Lower distances
and locality of traffic
• Better QoS for the User
UNIVERSITÀ DEGLI STUDI
ROMATRE
Related Work (Selection)• N. Chatzis, G. Smaragdakis, A. Feldmann, and W. Willinger,
“There is more to ixps than meets the eye”, SIGCOMM Comput. Commun. 2013
• A. Gupta, L. Vanbever, M. Shahbaz, S. P. Donovan, B. Schlinker, N. Feamster, J. Rexford, S. Shenker, R. Clark, and E. Katz-Bassett, “Sdx: A software defined internet exchange”,in Proc. SIGCOMM 2014
• M. Ahmad and R. Guha,“A tale of nine internet exchange points: Studying path latencies through major regional ixps”, in Local Computer Networks (LCN), 2012
UNIVERSITÀ DEGLI STUDI
ROMATRE
Motivation and Contribution
UNIVERSITÀ DEGLI STUDI
ROMATRE
Motivation and ContributionMOTIVATION
• A quantitative analysis on the impact of IXPs on key performance indicators is missing
• To our knowledge, all existing works use, at least partially, simulated data
• Comparing a path traversing an IXP with one that does not is hard without direct control of traffic flows
UNIVERSITÀ DEGLI STUDI
ROMATRE
Motivation and ContributionMOTIVATION
• A quantitative analysis on the impact of IXPs on key performance indicators is missing
• To our knowledge, all existing works use, at least partially, simulated data
• Comparing a path traversing an IXP with one that does not is hard without direct control of traffic flows
OUR CONTRIBUTION
• Supported by several ISPs, we compare key performance indicators using real-world measurements on IXP and non-IXP paths
UNIVERSITÀ DEGLI STUDI
ROMATRE
Side Motivation: DepeeringSome major ISPs have canceled all peerings (depeering) at the IXPs, justifying such decision in terms of :
• Cost reduction • Improvement of the quality of services (QoS)
UNIVERSITÀ DEGLI STUDI
ROMATRE
But…
UNIVERSITÀ DEGLI STUDI
ROMATRE
But…
UNIVERSITÀ DEGLI STUDI
ROMATRE
QoS InvestigationInvestigate the impact that IXPs have on the QoS
USED METRICS
UNIVERSITÀ DEGLI STUDI
ROMATRE
UNIVERSITÀ DEGLI STUDI
ROMATRE
QoS Investigation
Round Trip Time
Investigate the impact that IXPs have on the QoSUSED METRICS
UNIVERSITÀ DEGLI STUDI
ROMATRE
UNIVERSITÀ DEGLI STUDI
ROMATRE
Hop Count
QoS Investigation
Round Trip Time
Investigate the impact that IXPs have on the QoSUSED METRICS
UNIVERSITÀ DEGLI STUDI
ROMATRE
UNIVERSITÀ DEGLI STUDI
ROMATRE
Hop Count
QoS Investigation
Round Trip Time
Packet Loss
Investigate the impact that IXPs have on the QoSUSED METRICS
UNIVERSITÀ DEGLI STUDI
ROMATRE
UNIVERSITÀ DEGLI STUDI
ROMATRE
Jitter
Hop Count
QoS Investigation
Round Trip Time
Packet Loss
Investigate the impact that IXPs have on the QoSUSED METRICS
UNIVERSITÀ DEGLI STUDI
ROMATRE
UNIVERSITÀ DEGLI STUDI
ROMATRE
Traffic Locality InvestigationInvestigate the impact that IXPs have on keeping local the
local traffic
UNIVERSITÀ DEGLI STUDI
ROMATRE
A Case of StudyThe Italian Internet and the Italian main Internet eXchange Points
UNIVERSITÀ DEGLI STUDI
ROMATRE
A Case of StudyThe Italian Internet and the Italian main Internet eXchange Points
UNIVERSITÀ DEGLI STUDI
ROMATRE
A Case of StudyThe Italian Internet and the Italian main Internet eXchange Points
UNIVERSITÀ DEGLI STUDI
ROMATRE
A Case of StudyThe Italian Internet and the Italian main Internet eXchange Points
UNIVERSITÀ DEGLI STUDI
ROMATRE
Measuring SystemWe exploited the distributed infrastructure of RIPE Atlas, an Internet measurement network based on thousands of devices (probes), deployed all around the world.
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment CISCritical and most visited Internet Sites
Critical Internet Sites
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment CISCritical and most visited Internet Sites
Most visited Internet Sites
• We selected the sites that are most visited by italians according to the Alexa ranking
• We filtered the ones hosted in Italy
• Finally we selected the first 100
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment CIS: Setting
• Targets: • 50 critical Internet services • 100 most visited sites
• Duration of the experiment: 4 hours • Frequency of measurements:
• 6 pings per hour • 2 traceroutes per hour
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment CIS: ResultsMeasurement: Round Trip Delay
Targets: Critical Internet sites
0 %
20 %
40 %
60 %
80 %
100 %
0 20 40 60 80 100 120 140 160
Cu
mu
late
d f
ract
ion
of
pro
be
s
Round trip delay [ms]
Probes not traversing IXPsProbes traversing IXPs
UNIVERSITÀ DEGLI STUDI
ROMATRE
0 %
20 %
40 %
60 %
80 %
100 %
6 8 10 12 14 16 18 20 22 24
Cum
ula
ted fra
ctio
n o
f pro
bes
Hop count
Probes not traversing IXPsprobes traversing IXPs
Experiment CIS: ResultsMeasurement: Hop count
Targets: Most visited Internet sites
UNIVERSITÀ DEGLI STUDI
ROMATRE
0
20
40
60
80
100
120
AS174
AS3356
AS8928
AS3257
AS1200
AS2914
AS1299
AS6453
AS8220
AS3549
AS1273
AS20940
AS34419
% o
f tr
ave
rsin
g p
ath
s
UPSTREAMIXP
Experiment CIS: ResultsMeasurement: Paths traversing foreign (non-Italian) ASes
Targets: Critical Internet sites
UNIVERSITÀ DEGLI STUDI
ROMATRE
0
20
40
60
80
100
120
AS174
AS3356
AS2914
AS1299
AS3549
AS6939
% o
f tr
ave
rsin
g p
ath
s
UPSTREAMIXP
Experiment CIS: ResultsMeasurement: Paths traversing foreign (non-European) ASes
Targets: Most visited Internet sites
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment CIS: Discussion• IXPs have a positive effect on keeping local the local
traffic
• As a first observation, paths through IXPs seem to have better performance
• …However these measurements are performed by a variable set of probes, since we don’t have direct control on interdomain routing (i.e., BGP) announcements
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: SettingSelective BGP Announcements
UPSTREAMUPSTREAM
PARTNERISP
PHASES:UPSTREAM
IXP
ALL
Probe
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: SettingSelective BGP Announcements
UPSTREAMUPSTREAM
PARTNERISP
PHASES:UPSTREAM
IXP
ALL
Probe
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: SettingSelective BGP Announcements
UPSTREAMUPSTREAM
PARTNERISP
PHASES:UPSTREAM
IXP
ALL
Probe
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: SettingSelective BGP Announcements
UPSTREAMUPSTREAM
PARTNERISP
PHASES:UPSTREAM
IXP
ALL
Probe
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: SettingSelective BGP Announcements
UPSTREAMUPSTREAM
PARTNERISP
PHASES:UPSTREAM
IXP
ALL
Probe
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: SettingSelective BGP Announcements
UPSTREAMUPSTREAM
PARTNERISP
PHASES:UPSTREAM
IXP
ALL
Probe
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: SettingSelective BGP Announcements
UPSTREAMUPSTREAM
PARTNERISP
PHASES:UPSTREAM
IXP
ALL
Probe
UNIVERSITÀ DEGLI STUDI
ROMATRE
UPSTREAMUPSTREAM
PARTNERISP
PHASES:UPSTREAM
IXP
ALL
Probe
Experiment SBA: SettingSelective BGP Announcements
UNIVERSITÀ DEGLI STUDI
ROMATRE
Partner ISPsFor the second experiment we partnered with 3 providers:
Mc-Link
Italian network operator headquartered in Rome and founded in 1986.
It mainly offers services for the business
industry.
Seeweb
Italian network operator headquartered in Rome and founded in 1998.
It offers cloud services and web hosting
Unidata
Italian network operator headquartered in Rome and founded in 1985.
Among the co-founders of the popular IXP
NaMeX
UNIVERSITÀ DEGLI STUDI
ROMATRE
0 %
20 %
40 %
60 %
80 %
100 %
0 10 20 30 40 50 60 70 80
Cu
mu
late
d f
ract
ion
of
pro
be
s
Average round-trip delay [ms]
Probes choosing upstream (M)Probes choosing IXPs (M)
Probes choosing upstream (F)Probes choosing IXPs (F)
Experiment SBA: ResultsMeasurement: Average Round Trip Delay
Target(s): Inside Seeweb (“M” and “F”)
UNIVERSITÀ DEGLI STUDI
ROMATRE
0 %
20 %
40 %
60 %
80 %
100 %
0 10 20 30 40 50 60 70
Cum
ula
ted fra
ctio
n o
f pro
bes
Average round-trip delay [ms]
AS 174IXP
Experiment SBA: ResultsMeasurement: Average Round Trip Delay
Target: Inside Seeweb, through upstream AS 174
UNIVERSITÀ DEGLI STUDI
ROMATRE
0 %
20 %
40 %
60 %
80 %
100 %
2 4 6 8 10 12 14 16 18 20
Cum
ula
ted fra
ctio
n o
f pro
bes
Average hop count
Probes choosing upstreamProbes choosing IXPs
Experiment SBA: ResultsMeasurement: Average Hop count
Target: Inside MC-LINK
UNIVERSITÀ DEGLI STUDI
ROMATRE
0 %
20 %
40 %
60 %
80 %
100 %
5 6 7 8 9 10 11 12 13
Cum
ula
ted fra
ctio
n o
f pro
bes
Average hop count
AS 3356IXP
Experiment SBA: ResultsMeasurement: Average Hop count
Target: Inside MCLINK, through upstream AS 3356
UNIVERSITÀ DEGLI STUDI
ROMATRE
0 %
20 %
40 %
60 %
80 %
100 %
0 50 100 150 200 250
Cum
ula
ted fra
ctio
n o
f pro
bes
Jitter [ms]
Probes choosing upstreamProbes choosing IXPs
Experiment SBA: ResultsMeasurement: Jitter
Target: Inside Unidata
UNIVERSITÀ DEGLI STUDI
ROMATRE
0 %
20 %
40 %
60 %
80 %
100 %
0 50 100 150 200 250
Cu
mu
late
d f
ract
ion
of
pro
be
s
Jitter [ms]
AS 3257IXP
Experiment SBA: ResultsMeasurement: Jitter
Target: Inside Unidata, through upstream AS 3257
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: ResultsMeasurement: Gain on the RTT (per class of probes)
Target: Inside MC-LINK
0 %
20 %
40 %
60 %
80 %
100 %
-60 -50 -40 -30 -20 -10 0 10
Cu
mu
late
d f
ract
ion
of
pro
be
s
Gap in average round trip delay [ms]
CAHCustomer
Transit
UNIVERSITÀ DEGLI STUDI
ROMATRE
Experiment SBA: ResultsMeasurement: Gain on the Hop Count (per class of probes)
Target: Inside MC-LINK
0 %
20 %
40 %
60 %
80 %
100 %
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1
Cum
ula
ted fra
ctio
n o
f pro
bes
Gap in average hop count
CAHCustomer
Transit
UNIVERSITÀ DEGLI STUDI
ROMATRE
ConclusionsOur experiments put in evidence that peerings exploiting IXPs:
• have a positive effect on key performance indicators
• have the effect of reducing the number of foreign ISPs traversed to reach critical Internet services (keep local the local traffic)
UNIVERSITÀ DEGLI STUDI
ROMATRE
Future WorkIt would be interesting to:
• extend our experiments considering more performance indicators (e.g. bandwidth)
• cooperate with ISPs and IXPs in other countries to reproduce the experiments, compare the results and generalize the methodology