IXP Data Analysis
Anja Feldmann TU Berlin/T-Labs
Walter Willinger Niksun Inc
Nikos Chatzis George Smaradakis
Jan Böttger Nadi Sarrar
Thomas Krenc TU Berlin/T-Labs
Steve Uhlig Queen Mary
University of London
Bernhard Ager ETH Zürich
Internet: Mental model (before 2010)
http://conferences.sigcomm.org/sigcomm/2010/slides/S3Labovitz.pdf
Internet mental model – ala 2011
http://conferences.sigcomm.org/sigcomm/2010/slides/S3Labovitz.pdf
Google, Akamai,
RapidShare, …
Question – What about IXPs
Who are these IXPs?
What do they provide us with?
Google, Akamai,
RapidShare, … IXP
A bit of history about IXPs
Simple definition (to begin with):
“A physical point where Internet traffic can be exchanged”
North America
Decommissioning of National Science Foundation Network (NSFNET) around 1994/95
Establishment and operation of 4 Network Access Points (NAPs)
IXPs in US are the “descendents” of the NAPs From 4 in 1996 to more than 80 modern IXPs in US
Typically run as for-profit commercial entities worldwide in 2012
Fulfill NAPs„ orginal role:
• Provide physical infrastructure
• Carrier-neutral meeting place for ISPs to communicate with each other, independent of third parties
Internet eXchange Points (IXPs)
Layer-2 switch
AS4
Content
Provider 2
AS5
AS1
AS2 Content
Provider 1
AS3
IXPs Offer connectivity to ASes
A bit of history about IXPs (2)
NAP like facilities in Europe
Started in early 1990s
Motivation: Linking their networks for exchanging their local traffic locally vs. astronomical transatlantic bandwidth costs.
European IXPs:
European IXP marketplace has strived
Business model: Mainly not-for profit
Vibrant and innovative European IXP scene
> 150 IXPs from huge, to medium, to small
Infrastructure of an IXP (DE-CIX)
http://www.de-cix.net/about/topology/
Robust infrastructure with redundency
The role of IXPs today
Local traffic stays local
1. Reduced transit costs to upstream providers. Avoids paying the astronomical transatlantic bw costs
2. Improved network performance and QoS due to reduced delay (e.g., decreased round-trip times) and better routes (e.g., reduced number of AS hops for typical end-to-end paths)
3. Boost to local Internet economy due to “good” Internet
4. More international bw for expensive international traffic
Offer services for the “common good”
DE-CIX: A local power-house …
DE-CIX was founded in 1995 (one location in Frankfurt)
Started with 3 ISPs, is run by a non-profit organization
Today (April 2013), DE-CIX operates a distributed infrastructure, with 5 different locations within Frankfurt
Providing service at some 18 DCs/co-location facilities within the Frankfurt metro-area
Providing interconnection services at access speed from1 Gbps to multiple 100 Gbps ports
~500 members, ~15PB/day
IXP Traffic Growth, 2007-2012 (DE-CIX)
~10-20 PB/day early 2013
Outline
Introduction to IXPs
A large European IXP
IXP peering fabric
IXP member diversity
IXP traffic
IXPs as vantage points
Summary
Data – From collaboration with IXP
Collaboration with a major European IXP
Expanding… Started cooperations with two other IXPs
Data: Summary of traffic flows:
sFlow records
Recall: Petabytes per hour… • Sampled data: 1 out of 16 K packets
• Summaries: 128 bytes IP/TCP/UDP headers
Collected from 2011 onward
Data – From collaboration with IXP
Fact 1 – IXP members/participants
Apr 25 May 1
Aug 22 Aug 28
Oct 10 Oct 16
Nov 28 Dec 4
Member ASes 358 375 383 396
Tier-1 13 13 13 13
Tier-2 281 292 297 306
Leaf 64 70 73 77
Countries of member ASes 43 44 45 47
Continents of member ASes 3 3 3 3
Daily avg. volume (PB) 9.0 9.3 10.3 10.7
Traditional classification for 2012
Fact 2 – IXP members/participants
Member ASes often offer multiple services
By Business type
Fact 3 – IXP traffic
Traffic Volume: Same as Tier-1 ISPs
IXP is interchange for Tier-2 ISPs
Outline
Introduction to IXPs
A large European IXP
IXP traffic
IXPs as vantage points
IXP peering fabric
Summary
Fact 3 – traffic – top-10 tier-2 members
Pronounced time of day effects
Some ASes fully utilize their capacity
Application classification: Challenges
Classification methods
Port-based
Payload-based
Flow feature
Host-behavior
Limitation of sFlow data
Sampled sFlow records 1 out of 16K
Only first 128 bytes of each frame
For 83% (74%) of TCP (UDP) flows we see 1 packet
Neither method sufficient => use combination
Application classification: Approach
Use combined approach
Take advantage of UDP signatures
Application classification: Results
Application mix
Do different subsets of AS-links one “sees” different application mix
What is “representative”?
Identify end-point roles
There are 3 groups of /24s: client-only, server-only, and mixed
Outline
Introduction to IXPs
A large European IXP
IXP traffic matrix
IXPs as vantage points
IXP peering fabric
Summary
Are IXPs good vantage points? Yes!
What do we see? What do we miss?
N. Chatzis, G. Smaragdakis, J. Böttger, T. Krenc, A. Feldmann, and W. Willinger “On the benefits of
using a large IXP as an Internet vantage point” ACM IMC13
What about network heterogenity?
Focus on commercial traffic (Use server-related traffic)
Identify server IPs
Cluster them according to the organization responsible for the traffic
N. Chatzis, G. Smaragdakis, J. Böttger, T. Krenc, A. Feldmann, and W. Willinger “On the benefits of
using a large IXP as an Internet vantage point” ACM IMC13
A toy example
AS1
AS4
AS2
AS3
S
S
S
S
S
S
S
S
S
S
1. AS1 and AS5 belong to the same organization 2. AS2 hosts (servers of) many organizations 3. Owner of does not have an AS 4. AS3 is an IXP or part of ISP bundling
S
AS5 C
C C
C C C
S
AS6
AS7 C
C C
S
S C
Who are the main players?
AS1
AS4
AS2
AS3
S
S
S
S
S
S
S
S
S
S AS5 C
C C
C C C
S
AS6
AS7 C
C C
S
S C
1. Are they AS1-AS7 or the owners of ? Recall that these servers may be
“responsible” for ~2/3 of the traffic
… so their owner can shape the traffic!
S S S
S
Today‟s typical assumptions…
AS1
AS4
AS2
AS3
S
S S
S
S
S
S
S
S
S AS5 C
C C
C C C
S
AS6
AS7 C
C C
S
S C
1. So we cannot explain how
The owner of shapes the traffic
• Because we do not associate all its servers with it and we associate wrong servers with it
S
AS link heterogeneity
Given a member AS
How much traffic from, e.g., Akamai servers comes/goes from/to the Akamai member AS vs.
How much traffic from, e.g., Akamai servers comes/goes from/to non-Akamai member ASes
S
S
Akamai AS (member)
non-Akamai AS
(member) Akamai servers
member AS
Using IXPs as vantage points
The end of the traditional AS-level view of Internet
An AS-level view is outdated/not very informative
An end-to-end view provides more insight!
Outline
Introduction to IXPs
A large European IXP
IXP traffic
IXPs as vantage points
IXP peering fabric
Summary
IXPs – Publicly available information Sources: euro-ix, PCH, PeeringDB, IXP‟s sites
Generally known: # IXPs ~ 350 worldwide
http://www.pch.net
IXPs – Publicly available information
0
100
200
300
400
500
600
ASNs at IXP
Unique ASNs
https://www.euro-ix.net
Generally known: # IXPs ~ 350 worldwide
Somewhat known: # ASes per IXP up to 500
IXPs – Publicly available information
0
1000
2000
3000
4000
5000
6000
7000
Europe NorthAmerica
Asia/Pacific LatinAmerica
Africa
IXP Member ASes by region
https://www.euro-ix.net/tools/asn_search
Generally known: # IXPs ~ 350 worldwide
Somewhat known: # ASes per IXP up to 500
Less known: # ASes ~ 11,000 worldwide
IXPs – Publicly available information Generally known: # IXPs ~ 350 worldwide
Somewhat known: # ASes per IXP up to 500
Less known: # ASes ~ 11,000 worldwide
Even less known: IXPs =~ Tier-1 ISP traffic
AMS-IX
IXPs – Publicly available information Generally known: # IXPs ~ 350 worldwide
Somewhat known: # ASes per IXP up to 500
Less known: # ASes ~ 11,000 worldwide
Even less known: IXPs =~ Tier-1 ISP traffic
Unknown: # of peerings at IXPs
Peering links – 2012 estimates?
Methodology Number of peering links in the entire Internet
[Dhamdhere et al.] 2010 Lower bound estimate based on BGP data)
> 20,000
[Augustin et al., Chen et al.] 2009/2010 Targeted/opportunistic traceroute from network edge
> 40,000
Dasu 2011. Targeted data plane measurements
> 60,000
IXP peering link between pair of ASes if
IP traffic exchanged
• BGP traffic only (e.g., in case of backup links)
• IP otherwise
Potential links
Member ASes in Nov/Dec‟11: 396
396x395 / 2 = 78,210 P-P links possible
Observed links
> 50,000 peering links
Peering rate > 60%!
•
•
June‟12: 421
> 55,000 peering links!
Peering rate > 60%! > 60%!
Fact 4 – IXP peerings
Fact 4 – IXP peerings Internet-wide
Single IXP > 50,000 peering links
Derivation of new lower bound
10 large IXPs in Europe: ~160,000 peering links
Remaining 340 or so IXPs: ~ 40,000 peering links
Completely ignoring all other peerings
(Conservative) lower bound on #of peering links
> 200,000 peering links in today‟s Internet (as compared to currently assumed ~ 40,000 – 60,000)
Requires a revamping of the mental picture our community has about the AS-level Internet.
Fact 4 – IXP peerings Internet-wide
Methodology Number of peering links in the entire Internet
[Dhamdhere et al.] 2010 Lower bound estimate based on BGP data)
> 20,000
[Augustin et al., Chen et al.] 2009/2010 Targeted/opportunistic traceroute from network edge
> 40,000
Dasu 2011. Targeted data plane measurements
> 60,000
2012 (This talk) data from IXPs > 200,000
Summary
Large IXPs are ideal Internet vantage points
Internet monitoring is a „big data“ problem
Large IXP study reveals diverse eco-system wrt members, business types, connectivity, traffic, etc.
Large IXP supports rich peering fabric
Single IXP doubles the estimated number of peering links
Needs revamping of mental picture of AS-level Internet
Implications for studies of AS-level Internet
ASes – can no longer be treated as „homogeneous“
AS links – simple classification (peering, cust-prov) should fade
IXP peerings – when peering links are used as cust-prov links…
AS traffic – what traffic is carried by whom?