Transcript WP1 - TERENA

INTERMON
Advanced architecture for INTER-domain quality of service
MONitoring, modelling and visualisation
http://www.ist-intermon.org
U. Hofmann*, I.Miloucheva, T.Pfeiffenberger, F. Strohmeier
Salzburg Research,*Univ. of Applied Sciences, Salzburg Austria
{ulrich.hofmann,ilka.miloucheva,thomas.pfeiffenberger,felix.strohmeier}
@salzburgresearch.at
17/07/2015
Slide 1
1. INTERMON Overview
2. CM Toolset active QoS monitoring and analysis toolkit
- Motivation
- Components
- Scenario
- Results
3. Future Work
17/07/2015
Slide 2
1. Overview
Motivation : ISP wants to enhance the inter-domain Quality of Service
(QoS) analysis in large-scale, multi-domain to offer stabile inter-domain
services
Different
Destination AS
Transit AS
Different
Source AS
ISP T1
ISP D1
• ISP S1/2 are competitors
ISP D2
• ISP* offers SourceEndSystem
ISP S1
Source
end
system
ISP T2
ISP D3
ISP T3
Destination
end
system
• best_of {ISP T1/T2}
ISP S2
ISP D4
End-toend
QoS
•...ISP D*
Peering
possibility
Traffic study at
border router
Alternative
topologies
• the aggregate QoS
Endto-end
QoS
INTERMON toolkit
with integrated data
base
17/07/2015
Slide 3
=> scalable inter-domain QoS monitoring and analysis
effect of inter-domain routing and BGP-4 protocol behaviour
monitoring : IPFIX
modelling load <=> QoS : simulation ( fluid, hybrid
analytical M/G/*
visual data mining architecture
measurement based simulation technologies IPFIX => Simulation
pattern detection, outlier elimination
pattern compression PLA
using
common QoS database with
policy-controlled interworking of components
17/07/2015
Slide 4
QoS measurement and modelling
Active QoS
Passive
measurement
delay
and traffic flow
measureme
emulation
nt tool
(CM Toolset)
QoS Pattern
analyser
-outliers
-linear
approximation
„what if“
analysis
Traffic
matrix
MR
Collector
Traffic
flows
(IPFIX)
Analysis of
Traffic &
Topology
Impact on
End-to-end QoS
INTERMON DB:
- QoS param. Meas.
(delay, loss),
-SNMP data per router
-IPFIX traffic
-BGP-4 protocol data
-Traceroute topology
BGP-4
Protocol
Analyser
- Heuristics
for BGP-4
Patterns
Simulation
toolkit
-fluid
-time series
-hybrid
-analytical
Active
topology
(traceroute)
Analysis of
CM Toolset
17/07/2015
Slide 5
2. CM Toolset active QoS monitoring and analysis toolkit
Motivation
QoS offer S->D
source
ISP_B
ISP_A
destination
ISP_C
 QoS is different for different paths : s->A->B->d , s->A->C->d
Salzburg-Madrid
Dt=20ms
 QoS is different for different SLS : VoIP (160 Byte), FTP (1500 Byte)
17/07/2015
Slide 6
Components
GUI
(WWWBrowser)
GPS
Equipment
Web
Server
MDB
CMCaller
Measurement
Management Station
CMDaemon
Measurement Flows
(TCP/UDP)
Measurement
Client Station
IP Network
17/07/2015
GPS
Equipment
CMDaemon
Measurement
Client Station
Slide 7
Scenario:
detect router
anomaly
• define pattern
owd>2s & singular
• specify
measurements
rate, p_size,
duration,
source,
destination,..
CM Toolset Scenario for complex QoS analysis and
data mining
-capacity planning
-patterns dependent on topology change
-application emulation and QoS study - VoIP
Spatio temporal QoS Analyser
-Pattern detection (generic patterns for
data mining rules)
-Outlier
-Similarity detection
Pattern
data base
• measurements
• analysis
if(owd>2s &
singular)
then anomaly
else path_change
QoS measurement for
secified scenario
17/07/2015
QoS monitoring
data base
Slide 8
Results (1): Active Topology discovery using traceroute data base
 Objective: study of topology properties of the connection
 number hops,
 availability of the routers
 long term reporting : per hour, day
 BGP issues ( security,...)
Salzburg-Madrid
red: router not responding
brightness: RTT
blue: some RTT > e2e_RTT
 Separation of measurement results per path
17/07/2015
Slide 9
Results (2) Piecewise linear approximation
• data as sequences of straight lines,
• optimise the tradeoff between
• high level of aggregation of measurement results
• „pieces“: set of consecutive measurement points
with „similar“ gradient d_measurement(t)/dt
• plain : gradient =0
• increasing : gradient >0
• decreasing: gradient <0
• „similarity“ : tolerance parameter d
•Pattern Description Language (PDL)
17/07/2015
Slide 10
Salzburg-Sao Paulo: router anomaly ?
17/07/2015
Slide 11
3. Future Work
• improved outlier analysis e.g. prediction optimised (ARIMA)
• symbolic representation for pattern detection {R} -> {a,b,c,..}
• evaluation tradeoff : information loss, compression
• InterDomain Forum Cluster
• audio QoS pattern analysis ( ITU E.855, perceptual QoS )
• prototype for global controller ( GEANT ?)
• MOME cluster
• anomaly detection for network security
• monitoring and modelling for network planning
17/07/2015
Slide 12