Measuring the Spatial Structure of Traffic Congestion in the Internet Gábor Vattay Center for...

Measuring the Spatial Structure of Traffic Congestion in the Internet

Gábor Vattay

Center for Communication Networks Data Analysis, Collegium Budapest

Physics of Complex Systems, Eötvös University

ELTE-Ericsson Research Communication Networks Laboratory

Spatial Structure of Internet Traffic

The Blue Ocean • Telecommunication Services and Computer Hardware

Equipment is one of the most R&D intensive sectors (70 B$ R&D spending in 2005)

• Research spending traditionally goes into Electrical Engineering and Computer Science, Physics,

Complexity is not on the horizon due to cultural problems

• While in biology, economy/finance there is traditional scientific competition, telecommunication is still a competition free Blue Ocean for complexity research

• Interdisciplinary research of Engineering and Complex Systems has a great perspective if log-logging is avoided

Inspiration from complex system side: What you need?

Complex NetworkSelfsimilar Traffic

Nonlinearity and Chaos


Topology


The Faloutsos Graph 1999


Traffic


Typical internet traffic traces

W. E. Leland et al. SIGCOMM 93


Chaos


Chaos in Computer Networks


Periodicity

Veres & Boda INFOCOM 2000


Chaos



Ljapunov properties



Strange attractor



Active Internetmeasurements

• Internet: highly heterogeneous and decentralized

• why are we measuring state variables (e.g. loss-rates, delays, bandwith)?– to predict the quality of various services and applications over the

Internet,– to construct more efficient transfer protocols,– to analyze the spatial structure of the traffic, etc.

• active probing:– injecting probe packets + analyzing the received probe stream– end-to-end information about the participating nodes– resolved information: only with the cooperation of routers


Network Tomography

Method: we send back-to-back packet pairs and measure their end-to-end delay at arrival with very high precision

Goal: to resolve delay statistics on internal network segments too, where we do not have monitoring stations

Key idea: delay correlation on the common segment for the packets in a pair


Delay estimation for

the Y-topology Definitions:• N - number of successful pairs, where none of the probes is lost.• and , the end-to-end delay experienced by the

probes of the k-th pair

• Goal: Estimate the distribution of , and

from the end-to-end delays.

• Quantization of the delay into B bins of uniform size q

if ,

)(2 kY

)()()( 211 kXkXkY )()()( 312 kXkXkY Nk ..1

)(1 kX )(2 kX )(3 kX

)(1 kY

jqkX di )( jqkXqj i )()1( Bji ,..2,13,2,1


B

j ji

ji

n

n

jiP1 ,

,

,

)()()( 211 kXkXkY ddd

)()()( 312 kXkXkY ddd

1)(1)(1

)(,3)(,2,1

21

),(

),(

smBslBsBH

smslHs

s

dd

PPPmlP

mqYlqYP

))(),(( 211

nYnYPL ddN

n


EM-algorithm

B

j ji

ji

n

n

jiP1 ,

,

,

),(),|(

),(),|(

),(),|(

,3)(,2)(,13

)(,3,2)(,12

)(,3)(,2,11

mlP

PPPmqlqjqXP

mlP

PPPmqlqjqXP

mlP

PPPmqlqjqXP

jjmljmd

jlmjjld

jmjljd

))(),(|)(( 211

, mYkYjqkXPn dddi

N

kji

0, jiP


msX 1832 msX 2283

msX 1751


Real and virtual measurement

points

The etomic active probing infrastructure

Spatial Structure of Internet Traffic Best Testbed Award

• The European Traffic Observatory Measurement InfrastruCture (etomic) was created in 2004-05 within the Evergrow Integrated Project launched by the Future and Emergent Technologies Programme of the European Union.

• Its goals:– to provide an open access, public test bed for

researchers investigating the Internet with active measurement methods

– to serve as a Virtual Observatory active measurement data on the European part of the Internet

History

http://www.tridentcom.org/tridentcom05/bta05.htm


Founders• Its Central Management System (CMS) has been

developed by the Grupo de Redes, Sistemas y Servicios Telemáticos Departamento de Automática y Computación Universidad Pública de Navarra.

• Its hardware infrastructure has been designed and built by the Cooperative Center for Communication Network Data Analysis in Collegium Budapest Institute for Advanced Study.

• The measurement stations are hosted by European research groups collaborating in the Evergrow project.


Measurement sites


Visualization

2005.07.01. 3:30AM


Topology of routers

etomic stationsrouters


Snapshot of queueing dealys in Europe


Daily change of mean queuing delays


Distribution of the mean queuing delay

(night 3:30)



(afternoon 16:00)



(day 14:00)

Variance vs. mean

Day

Night

Afternoon


Main results

Log-normal distribution of the delay Variance and delay are proportional

Future measuremnets


Growing number of monitored links


The DIMES project


DIMES Agents in Europe


Measurement sites

Properly located Dimes agents (red)Branching routers (blue)


Expectations 2.

Number of Dimes agents

Num

ber

of d

isco

vere

dse

gmen

ts/b

ranc

hing

rou

ters

Newly discovered internal segments

Newly discovered branching routers

Traceroute between 257 Dimes agents and 15 Etomic nodes


Thanks!

IST Future and Emerging Technologies

Measuring the Spatial Structure of Traffic Congestion in the Internet Gábor Vattay Center for...

Documents

Transcript of Measuring the Spatial Structure of Traffic Congestion in the Internet Gábor Vattay Center for...