1 Optimal Multi-Path Routing and Bandwidth Allocation under Utility Max-Min Fairness Jerry Chou and Bill Lin University of California, San Diego IEEE IWQoS.

1

Optimal Multi-Path Routing and Bandwidth Allocation under Utility Max-Min Fairness

Jerry Chou and Bill LinUniversity of California, San Diego

IEEE IWQoS 2009Charleston, South CarolinaJuly 13-15, 2009

2

Outline

• Problem

• Approach

• Application to optical circuit provisioning

• Summary

3

Basic Max-Min Fair Allocation Problem

• Motivation: Bandwidth allocation is a common problem in several network applications

• Example:

A D

10 10

B1010

C

C1: AD C2: BD C3: CD

C1 C2 C3

5

Maxincrease

SaturatedflowsFully allocated link

4

Utility Max-Min Fairness

100/)( 21 rr 100/)12()( 2

2 rrr 100/)403()(3 rrC1: AD C2: BD C3: CD

0 BW 10

utility

1

0

utility

1

0

utility1

00 BW 10 0 BW 10

Path of C1 Allocation Utilities

ABD (5, 5, 10) (0.25, 0.85, 0.70)

Utility functions capture differences in benefitsfor different commodities

A D

10 10

B1010

C

5

Utility Max-Min Fairness

100/)( 21 rr 100/)12()( 2

2 rrr 100/)403()(3 rrC1: AD C2: BD C3: CD

0 BW 10

utility

1

0

utility

1

0

utility1

00 BW 10 0 BW 10

A D

10 10

B1010

C

Path of C1 Allocation Utilities

ABD (5, 5, 10) (0.25, 0.85, 0.70)

ABD (6.8, 3.2, 10) (0.47, 0.47, 0.70)

Utility functions capture differences in benefitsfor different commodities

6

Utility Max-Min Fairness

100/)( 21 rr 100/)12()( 2

2 rrr 100/)403()(3 rrC1: AD C2: BD C3: CD

0 BW 10

utility

1

0

utility

1

0

utility1

00 BW 10 0 BW 10

A D

10 10

B1010

C

Path of C1 Allocation Utilities

ABD (5, 5, 10) (0.25, 0. 85, 0.70)

ABD (6.8, 3.2, 10) (0.47, 0.47, 0.70)

Multi-path (8, 4, 8) (0.64, 0.64, 0.64)

6

2

Freedom of choosing multi-path routing achieveshigher min utility and more fair allocation

7

Prior Work

• Utility max-min fair allocation only considered fixed (single-path) routing

• Optimal multi-path routing only considered weighted max-min and max-min fairness

8

Why is the Problem Difficult?

• Why is optimal multi-path routing and allocation under utility max-min fairness difficult?

→ Unlike conventional fixed (single) path max-min fair allocation problems1. Cannot assume a commodity is saturated just

because a link that it occupies in the current routing is full

2. Once a commodity is saturated, cannot assume its routing is fixed in subsequent iterations

9

Example

• At iteration i, suppose we route both flows AD and AE with 5 units of demand

AD:5

AE:5

If routing is fixed after iteration, AD would be at most 5

A D

10/10 5/10

B

0/10

CE

5/5

0/10

10

Example

• At iteration i+1, suppose we want to route AD with 10 units of demand

AD:10

AE:5

Route of AD must change to increase

A D

5/10 0/10

B

10/10

CE

5/5

10/10

11

Outline

• Problem

• Approach– OPT_MP_UMMF– ε-OPT_MP_UMMF

• Application to optical circuit provisioning

• Summary

12

OPT_MP_UMMF

• Step 1: Find maximum common utility that can be achieved by all unsaturated commodities

• Step 2: Identify newly saturated commodities

• Step 3: Assign the utility and allocation for each newly saturated commodity

13

Key Differences

• A commodity is truly saturated only if its utility cannot be increased by any feasible routing – Requires testing each commodity for saturation

separately

• To guarantee optimality, fix the utility, not the routing after each iteration

Fix utility,not routing

14

Comments• Although OPT_MP_UMMF achieves optimal

solution, both Steps 1 & 2 require solving non-linear optimization problems

Step 1 Step 2

15

ε-OPT_MP_UMMF

• Instead of solving a non-linear optimization problem, find maximum common utility by means of binary search

• Test if a common utility has feasible multi-path routing by solving a Maximum Concurrent Flow (MCF) problem

16

Maximum Concurrent Flow (MCF)

• Given network graph with link capacities and a traffic demand matrix T, find multi-path routing that can satisfy largest common multiple of T

• If < 1, means demand matrix cannot be satisfied

• If > 1, means bandwidth allocation can handle more traffic than specified demand matrix

• MCF well-studied with fast solvers

17

Find Maximum Utility• Determine demand matrix by utility functions• Find feasible routing by querying MCF solver

– If <1, decrease utility, otherwise increase utility

20

2010 30 40 50

4060

80

100

20

2010 30 40 50

4060

80

100

20

2010 30 40 50

4060

80

100

20

2010 30 40 50

4060

80

100

BW BW BW BW

Utility(%

)

Utility(%

)

Utility(%

)

Utility(%

)

C = 100Max utility Traffic (T)

1 (50,50,50,50) 0.5

.0.6±ε (10,40,10,40) 1

0.5 (10,30,10,40) 1.25

18

Outline

• Problem

• Approach

• Application to optical circuit provisioning

• Summary

19

Optical Circuit Provisioning Application• Provision optical circuits for Ingress-Egress (IE) pairs

to carry aggregate traffic between them

• Goal is to maximize likelihood of having sufficient circuit capacity to carry traffic

WDM links

Optical circuit-switched long-haul backbone cloud

Boundaryrouters

Opticalcircuit switches

20

Optical Circuit Provisioning (cont’d)• Utility curves are Cumulative Distribution Functions

(CDFs) of “Historical Traffic Measurements”

• Maximizing likelihood of sufficient capacity by maximizing utility functions

• Route traffic over provisioned circuits by default

• Adaptively re-route excess traffic over circuits with spare capacity

• Details can be found in– Jerry Chou, Bill Lin, “Coarse Circuit Switching by Default,

Re-Routing over Circuits for Adaptation”, Journal of Optical Networking, vol. 8, no. 1, Jan 2009

21

Experimental Setup

• Abilene network– Public academic network– 11 nodes, 14 links (10 Gb/s)

• Historical traffic measurements– 03/01/4 – 04/21/04

22

Example

Seattle

Sunnyvale

Indianapolis

Denver

Los Angeles Kansas City

Chicago New York

Washington

Atlanta

Houston

SeattleNY:90% time ≤ 6Gb/s50% time ≤ 4Gb/sAllocate: 6Gb/s

SunnyvaleHouston:90% time ≤ 6Gb/s80% time ≤ 4Gb/sAllocate: 4Gb/s

Seattle NY has 90% acceptance probabilitySunnyvale Houston has 80% acceptance probability

23

Comparison of Allocation Algorithms• WMMF: Single-path weighted max-min fair

allocation– Use historical averages as weights– Only consider OSPF path

• UMMF: Single-path utility max-min fair allocation– Only consider OSPF path

• MP_UMMF: Multi-path utility max-min fair allocation– Computed by our algorithm

24

Individual Utility Comparison• Reduce link capacity to 1 Gb/s• MP_UMMF has higher utility for most flows

25

Minimum Utility Comparison • MP_UMMF has greater minimum utility

improvement under more congested network

26

Excess Demand Comparison

• Simulate traffic from 4/22/04-4/26/04• MP_UMMF has much less excess demand

27

Summary of Contributions

• Defined multi-path utility max-min fair bandwidth allocation problem

• Provided algorithms to achieve provably optimal bandwidth allocation

• Demonstrated application to optical circuit provisioning

28

Thank You