Distributed Policy-based Management of Measurement-
based Traffic Engineering
Steven Van den Berghe
IMEC / Ghent University (Intec)
outline
draft-ietf-tewg-measure
Measurement-basedTunnel management
2000
2002
2004
?
Tequila (http://www.ist-tequila.org)
SLS management
Traffic Engineering
Data Plane
Monitoring
Policy Management
PolicyConsumer
Pol. Mgttool
SLS Subs
SLS invoc.
TrafficForecast
DRsM
DRtM
MPLSTC
ND
SLS Repos.
PHB
Tequila (http://www.ist-tequila.org)
SLS management Traffic
Engineering
Data PlaneMonitoring
Policy Management
USER
SLSDemand
Resources
Status Feedback
Configuration
Resource Provisioning
2-level TE
DiffServ / MPLS config
Admission Control
Path Selection
SLSsubscription
Demand Forecast
SLSInvocation
InjectedTraffic
Requirements Feedback Mechanism
Operating in Multipath / Multiclass environment (DiffServ over MPLS)Large set of possible metrics One way Loss, delay, offered load
Sollicited / Unsollicited report triggering Regular status updates If (delay>x) then trigger
Need for measurement architecture
Enduser-to-Enduser
Core
Customer Customer ISP Aggregate Flow
B C
B
C
C
C
C
Ingress-to-Egress
Where / How to Measure
Active MeasurementEndpoints
Passive Measurementpoint
Scalability/Multipath IssueEvery ingress 2 every egress 4 every CoSNo single path => no single measurement
result
Complexity IssueNew measurement functions needed in every node
Could also be an
ISP
Trust IssueIsn’t nobody touching my injected packets
Not for TE
SLS Monitor
Network Monitor
Measurements For TE: tequila approach
I C E
Node Monitor
Node Monitor
Node Monitor
AM AM
PM
PM
PM
AM
SLS scope
(Link,PHB) scope (Link,PHB) scope
What to Measure
Service Level Agreement (SLA)Transport Service
Service Level Specification (SLS)
QoS class / PathPer Domain Behaviour (PDB)
Per Hop Behaviour (PHB)Traffic Conditioning Block
Scheduler (e.g. WFQ)Algorithmic Dropper (e.g. RED)
Throughput Monitoring
End-to-End delay/lossLSP stats
Per Hop Delay/Loss/Queue
stats
Down at the wire
Edge
Classifier
Classifier
Conditioning
Classifier
Edge
SelectionDecision
Classifier
Core
LSP Group
LSP
LSP
Queue LossOffered Load
LSP Offered Load
ClassifiedLoad
Conditioning Loss
LSP Throughput
ClassifiedLoad
Active Measurement Loss/DelayHow do I manage
all of this ?
Measurement-Based Tunnel Management
2-Level Traffic Engineering High-level multipath balancing &
DiffServ provisioning Low-level Path Selection based on
accurate network status
TunnelManagement
DiffServManagement
Uniform MeasurementManagement
TunnelManagement
Uniform MeasurementManagement
Case study: Level 2 Tunnel Management
TunnelManager
Evaluator
Monitor
If signal then switch traffic to Path with less problems
hx
hxllh
lxlx
xp
1
0
)(h
l
Signal:
Impact of Read-Out Periods
0
0.5
1
1.5
2
2.5
11 users 13 users 15 users
Loss
(%
)
Simulation Readout 0.25
Simulation Readout 0.5
Simulation Readout 1
Simulation Readout 2.5
TunnelManager
Monitor
Evaluator
RSVP-TE
Active vs. Passive
0
10
20
30
40
50
60
70
80
90
18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34Time (read-out 1 second)
Loss (%)
Passive Active Probe Every 0.2
Active Probe Every 0.5 Active Probe Every 0.7
20 applicationsstarted & buffers
full
Active loss measurements
Long RO needed to be accurate
Long RO-> less responsive
Try to Avoid Active Measurements
What’s Next
TunnelManagement ‘Tunnel’ creation,
maintenance & traffic mapping
Regardless of technologyRegardless of application
MPLSIPsecP2PGrid
TunnelSet-up
TunnelMonitoring
TunnelMapping
What’s Next
TunnelManagement
Uniform MeasurementManagement
Uniform access to measurements & reporting
Regardless of technologyRegardless of application
Next to TE: DoS, network diagnostics, accounting,….
Resource control
Aggregation close to the wire (~high speed links)
Security
Thank You
Any Questions ?
http://www.ist-tequila.org
http://www.ist-scampi.org
http://www.ietf.org/internet-drafts/
draft-ietf-tewg-measure-02.txt