Active Measurement Project on KREONET & APAN

2002. 1. 24Manhee Lee

mhlee@kreonet2.netSupercomputing Center

Korea Institute of Science & Technology Information

Objectives of this presentation

Introduction to KREONETNetwork Analysis Infrastructure of

KREONETIntroduction to AMPInteresting results of AMPProposal to make AMP APAN

Introduction to KREONET

Korea Research Environment Open Network

one of the Korea’s national core networks since 1988

Providing an advanced S&T and R&D infrastructure to 300 R&D institutes and universities

KREONet2, next generation research network, NGI(Next Generation Internet) technologies through interconnection with STAR TAP

Help researchers process large capacity of data in supercomputers and provide activation of world-class collaborative application service

Role of KREONET

High-Tech S&T devices(Supercomputers, CAVE,DB Servers…)

High-Tech S&T applications based KREONET/KREONet2/HPCNet

Why Network Measurement?

Network Operators’ View Is the network reliable?How is the network used?When should the network be upgraded?

Network Users’ ViewHow much do I use the network?Should I invest more or not?Does the Service Provider abide by its agreement?

Active Measurement

Method inject measurement traffic into the network

MetricsRound Trip Time, Packet loss, Topology

ToolsPing, Traceroute, NIMI, Surveyor, PingER, AMP

Passive Measurement

MethodDo not inject measurement traffic but just observe

traffic passivelyMetrics

Link Utilization, Traffic amount, Traffic classification, Traffic pattern

ToolsTcpdump, MRTG, Flowscan, OC3MON(CoralReef), F

lowscan+

Network Analysis Infrastructure(NAI)for KREONET

ATM Switch

KOREN/KREONet2 Router

KOREN/KREONet2-STAR TAP

KREONET Router

STAR TAP

FlowScan+ OC3MON

AMP

PingER

Netflow

AMP

AMP

AMP

PingER Destination hosts

Passive Measurement

Active Measurement

MRTG

AMP (Active Measurement Project)

Developed by NLANR/MOAT(SDSC)A distributed network of over 120 active

monitorsSystematically performing scheduled

measurements(fping, traceroute) between each other monitor

All monitors send data to central data collector periodically located in SDSC

Raw and analysis data available by Web

What can AMP tell us?

Round trip time once a minuteTraces between monitors every 10

minuteDistribution of RTTPacket LossHistory of changes of tracesYearly, weekly, daily RTT graph

Deployment Map of AMP U.S

Structure of AMP

Analysis machine

active monitor

active monitor

Other target

Analysis machine

Cichlid3D

Web Browser

active monitor

AMP Architecture

System cron process

fpingtraceroute

AMPlets

pingtraceroute

ping resultstraceroute results

am_master

am_slave

ping resultstraceroute results

apache

cgi-bin

trigger

File read

TCP connection

ping resultstraceroute resultsping results

traceroute results

File writeFile read

am_slave

AMPlet1

AMPlet2

AMPlet3

am_slave TCP connection

TCP connection

File write

HTTPRequest

WebService AMP Server

pingtraceroute

How to use AMP?

First, select an interested AMP monitor

Currently, there are over 120 monitors

For example, let’s click on amp-arizona

You can see brief traffic analysis from amp-arizona to all other monitors

yesterday’s minimum, mean, and max RTT, and packet loss rate

If you want to know specific traffic analysis, click the interested monitor

For example, let’s click on Harvard University

How to use AMP?

You can see yearly and weekly RTT graphs from Arizona State Univ. to Harvard Univ.

If you want to know more details about a specific day, click any date

For example, let’s click on 17 Dec 2001

How to use AMP?

You can see more details of the day from Arizona to Harvard

Daily RTT : around 62ms Hop count: 11 Jitter: around 1ms RTT distribution: mostly

62ms packet loss: many times,

10%

How to use AMP?

If click route, you can see all routes of the day

If there is any changes, you can see all the changes

How to use AMP?

amp-korea

NLANR/MOAT(SDSC) gave a monitor to KISTI and allowed it to be joined the AMP infrastructure for international collaboration.

amp-korea is monitoring the performance from Korea to about 120 amp U.S sites via STAR TAP.

Those amp U.S sites are doing so in the same way.

http://watt.nlanr.net//active/

http://watt.nlanr.net//active/

amp-korea HPC results

Routing Anomaly at 2001.8.14

Routing Anomaly at 2001.8.14

Problem Analysis in 2001.9

From amp-korea to amp-ncsaRTT anomaly detected. Why?

Problem Analysis in 2001.9

First compare with other network.From amp-korea to amp-sdscNearly same results.

Problem Analysis in 2001.9

Compare with other network.From amp-korea to amp-ucsdTotally different resultsWe can assume that this problem may not be

KOREN/KREONet2-STAP TAP link problem but STAR TAP network problem.

Problem Analysis in 2001.9

Take a look at STAR TAP network.From amp-startap to amp-ncsaAt normal case, its RTT is about 2,3 msWhy? Are there any routing changes?

Problem Analysis in 2001.9Take a look at its routing, but cleanThen, we can assume that the reason is

not routing problem but may be congestion or router’s low performance

Problem Analysis in 2001.9 Compare other network from STAR TAP Then, we can know that all traffic patterns STAR TAP t

o vBNS, Abilene, and CA*net are same So we can conclude that the problem of low network p

erformance might be in STAR TAP routers or inside network.

AMP KREONET

To build an AMP infrastructure in KREONET Last year, we set up two servers and 4 test AMP monit

ors at KREONET NOC This year, we plan to install dozens of AMP monitors in

KREONET network centers and major supercomputer users.

This AMP infrastructure will monitor the performance of the KREONET

URL: http://matthew.kreonet2.net (under construction) Software and Server Installing Guide :

http://watt.nlanr.net/

Proposal to make AMP APAN

APAN’s individual networks, institutes, and universities have been doing their own measurements

But there is no APAN-wide or Asia-wide measurement infrastructure

Because of increasing collaboration within Asia, it will be very useful to actively measure network performance of APAN

AMP may be a good candidate for APAN performance measurement infrastructure

Expected benefits from AMP APAN

RTT, Packet loss, connectivity among APAN countries

Same information with U.S and other countries by peering with AMP U.S

Optional performance test for APAN Performance testbed for future project

such as End-to-end Performance Initiative in APAN

Is it difficult to join or make AMP APAN?

Absolutely NOT!!!!!!!!!!! Making AMP APAN is very easy because it is cheap an

d easy to install You just need one FreeBSD system Please send us an email any time, then we can help y

ou install software Easy to join, easy to collaborate, easy to get informati

on and cheap to start!! AMP APAN will be very useful for APAN Why don’t you join us now?

Conclusions

With AMP, we can know RTT, Packet loss, traces, and throughput(optional)

AMP APAN will be very useful measurement infrastructure

Just one FreeBSD system is needed.Small Investment, Big Benefit!!

Links and Contacts

Links NLANR: www.nlanr.net AMP: amp.nlanr.net amp-korea: watt.nlanr.net//active AMP KREONET: matthew.kreonet2.net (under construction)

Contacts Engineer: Manhee Lee mhlee@kreonet2.net Director: Okhwan Byeon ohbyeon@kreonet2.net AMP US director: Ronn Ritke rritke@nlanr.net Developer: Tony McGregor tonym@cs.waikato.ac.nz

Comments or Questions?

Thanks…