iREX : Efficient Interdomain Resource Exchange Architecture

1
iREX Efficient Inter-domain Resource Exchange
Architecture
Globecom 2006

• Ariffin Datuk Yahaya, Tobias Harks Tatsuya Suda
• ariffin_at_ics.uci.edu harks_at_zib.de
  suda_at_ics.uci.edu

University of California, Irvine
Zuse Institute, Berlin
2
iREX Presentation Outline

• Introduction
• Background Inter-domain QoS Policy
• Problem Statement
• Our Solution
• Current Focus Efficiency
• Conclusion

3
Background Inter-domain (ID) QoS Policy

• Defines how a domains resources are allocated to
  honor the special needs of QoS traffic (i.e.
  video, voice, etc).
• Produces the effect of
• cooperation
• among multiple heterogeneous domains
• aggregated QoS service behavior
• from source to destination (end to end)

D
Provide Qos To Traffic x From S to D

ATT
SBC
MCI
INTERNET


Level 3
QWEST
Domains
Border Router
ATT has deployed ID QoS policy from S to D
through SBC and QWEST
4
Problem Statement

• How to automate inter-domain QoS policy?
• Issue
• Domains are heterogeneously managed and
  non-cooperative
• Hard to automate
• automation requires cooperation

5
iREX Presentation Outline

• Introduction
• Our Solution
• Goals and Approach
• Economic Market
• Current Focus Efficiency
• Conclusion

6
iREX
Inter-domain Resource Exchange Architecture

• Goals
• Self Manage QoS Policy Negotiation Deployment

• Approach
• Enable an economic market for inter-domain
  network resources
• Domains optimize for their own benefit
• Ensure architecture conformance
• Domain Reputation System
• Fault Tolerance System

Not presented
7
iREX Presentation Outline

• Introduction
• Our Solution
• Goals and Approach
• Economic Market
• Overview
• Example
• Current Focus Efficiency
• Conclusion

8
iREX use of Economics

• iREX Economic Model Posted Price Competition
• Domains are both sellers and buyers.
• Sellers
• Advertise the price of their available
  inter-domain resources
• Buyers
• Choose among advertised resources to deploy
  inter-domain QoS policy

Need QoS to ATT
Path A 20
Path C 5
Path B 10
9
iREX Presentation Outline

• Introduction
• Our Solution
• Goals and Approach
• Economic Market
• Overview
• Example
• Note Congestion Avoidance Benefit
• Research Question
• Current Focus Efficiency
• Conclusion

10
Economic Market Example
W
X
D
Path Vector at Y
S
Y
Z
Each resource owner decides on a price for his
links and periodically advertises the cheapest
known link prices. Domains form path vectors with
metrics based on price.
11
Economic Market Example
W
X
D
Path Vector at Y
S
Y
Z
After a few iterations domains have a path vector
of the cheapest priced resources to all
destinations.
12
Economic Market Example
1
W
X
D
1
SYWXD4
1
S
Y
Z
1
Y sends S its latest path vector and S gets
information that SYWXD is the cheapest path.
S then uses source routing to deploy QoS policy
along SYWXD.
13
Economic Market Example
2
W
X
D
2
SYWXD8
2
2
2
S
Y
Z
2
Bandwidth use causes link prices to increase.
14
Economic Market Example
3
W
X
D
3
SYWXD12
3
2
2
S
Y
Z
3
S continues to use the same path for new
deployments and prices continue to increase.
Domains continuously advertise the cheapest
prices and S gets new information on a cheaper
route SYZXD.
15
Economic Market Example
3
W
X
D
3
SYZXD10
Multiple QoS Policy Deployments Active On 2 paths
3
2
2
S
Y
Z
3
New Successful Deployment
S chooses SYZXD for next deployment(s).
16
iREX Presentation Outline

• Introduction
• Our Solution
• Goals and Approach
• Economic Market
• Overview
• Example
• Note Congestion Avoidance Benefit
• Research Question
• Current Focus Efficiency
• Conclusion

17
Congestion Avoidance Benefit
4
W
X
D
3
Sending simultaneously on more than 1 path ?
QoS Policy Deployments Active On 2 ID paths
3
3
3
S
Y
Z
4
Deployments avoid congestion automatically
by distributing deployments.
18
iREX Presentation Outline

• Introduction
• Our Solution
• Current Focus Efficiency
• Approach
• Results
• Conclusion

19

• Research Question
• How efficiently does iREX actually distribute
  network load?

iREX Causes Less Congestion
Congestion Results (INFOCOM 2006)
20
Approach
iREX protocol
Online
iREX Simulator
iREX SLA Time t Network Congestion
Generated Requirements
Efficiency Loss Metric
Compare Time t Snapshots
Domain Requirements
Domain Requirements
t
Domain Requirements
Time t Domain Requirements
Math Model
SNAPSHOT Requirements
Best Case Time t Network Congestion
Offline
CPLEX Math Optimizer
21
Math Model 1 (MinCONG)

• Represents the best case bandwidth usage scenario
• Concept
• Assign a tax to links that use bandwidth capacity
• Optimization Goal

22
Math Model 2 (MinMAX)

• Represents the best case bandwidth usage scenario
• Concept
• Minimize the most congested link
• Optimization goal

23
iREX Presentation Outline

• Introduction
• Our Solution
• Current Focus Efficiency
• Approach
• Results
• Conclusion

24
Metric Efficiency Loss

• Description
• How close iREX is to the Best Case
• Math solution for splitting reservations on
  multiple paths.
• Note iREX currently uses single path.
• Purpose
• Show how far iREX is from a mathematical solution.

25
iREX MINCONG
iREX MINMAX
SLA MINCONG
SLA MINMAX
26
Main Observation

• Efficiency Loss for iREX is low in the important
  region
• Less than 30
• Compare to SLA 150 to 350

27
iREX Presentation Outline

• Introduction
• Our Solution
• Current Focus Efficiency
• Conclusion

28
Conclusion

• iREX
• Efficient architecture for self-managing
  inter-domain QoS policy deployment
• Ideas
• Domains cooperate indirectly by optimizing their
  own benefit
• Price determines inter-domain path dynamics ?
  efficiency

29
Thank you.

• Contact Ariffin Datuk Yahaya
• ariffin_at_ics.uci.edu

30
Why does Resource Price increase?
Peak 1.77

• Statistical Multiplexing ? Risk
• Domains sell more resource than is actually
  available.
• Hope that traffic spikes do not occur at the same
  time
• Risk
• QoS degradation
• Lost data
• As incoming reservations increases
• Domain has more risk
• Increases price
• Willingness to pay also increases
• Incentive for domain to increase price

Peak 1.67

Peak 1.86
Non-Multiplexed peak total 5.3
Cumulative Peak 3.92
Resource
Time
Statistical GAIN 26