[PPT] - An Adaptive Multi-Temporal Approach for Robust Routing Pedro CASAS PowerPoint Presentation

SLIDE 1

EuroFGI Workshop on IP QoS and Traffic Control

An Adaptive Multi-Temporal Approach for Robust Routing

Pedro CASAS & Sandrine VATON

ENST Bretagne

EuroFGI Workshop on IP QoS and Traffic Control Lisbon, Portugal, December 6-7 2007

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 2

EuroFGI Workshop on IP QoS and Traffic Control

Traffic Engineering for Routing Optimization under Traffic Uncertainty Robust TE Techniques to tackle the problem ⇒ Robust Routing Stable (static) routing to avoid potential instabilities but.....

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 3

EuroFGI Workshop on IP QoS and Traffic Control

Traffic Engineering for Routing Optimization under Traffic Uncertainty Robust TE Techniques to tackle the problem ⇒ Robust Routing Stable (static) routing to avoid potential instabilities but..... NOTHING COMES FOR FREE ! !

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 4

EuroFGI Workshop on IP QoS and Traffic Control

Outline

1

Traffic Engineering (TE) under traffic uncertainty

2

Robust and proactive TE : the Stable Robust Routing

3

A time-varying approach : the Multi-Temporal Robust Routing

4

Conclusions and Perspectives

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 5

EuroFGI Workshop on IP QoS and Traffic Control

Outline

1

Traffic Engineering (TE) under traffic uncertainty

2

Robust and proactive TE : the Stable Robust Routing

3

A time-varying approach : the Multi-Temporal Robust Routing

4

Conclusions and Perspectives

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 6

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

Unexpected Events

External Routing Changes Equipment Failures Network Attacks Flash Crowds Spontaneous Services (P2P)

Sources of Demands Variation

Daily Periodic Usage Patterns

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 7

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

Unexpected Events

External Routing Changes Equipment Failures Network Attacks Flash Crowds Spontaneous Services (P2P)

Sources of Demands Variation

Daily Periodic Usage Patterns

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 8

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

External Routing Changes Equipment Failures Network Attacks Flash Crowds Spontaneous Services (P2P)

Sources of Demands Variation

Daily Periodic Usage Patterns Unexpected Events

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 9

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

Equipment Failures

Network Attacks Flash Crowds Spontaneous Services (P2P)

Sources of Demands Variation

Daily Periodic Usage Patterns Unexpected Events

External Routing Changes

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 10

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

Network Attacks Flash Crowds Spontaneous Services (P2P)

Sources of Demands Variation

Daily Periodic Usage Patterns Unexpected Events

Equipment Failures External Routing Changes

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 11

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

Flash Crowds Spontaneous Services (P2P)

Sources of Demands Variation

Daily Periodic Usage Patterns Unexpected Events

Network Attacks

Equipment Failures External Routing Changes

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 12

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

Spontaneous Services (P2P)

Sources of Demands Variation

Daily Periodic Usage Patterns Unexpected Events

Flash Crowds Network Attacks

Equipment Failures External Routing Changes

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 13

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

Spontaneous Services (P2P) Flash Crowds Network Attacks

Equipment Failures External Routing Changes

Sources of Demands Variation

Daily Periodic Usage Patterns Unexpected Events

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 14

EuroFGI Workshop on IP QoS and Traffic Control

TE in current scenario : a challenging task

Current traffic demands are highly variable and uncertain

Sources of Demands Variation

Daily Periodic Usage Patterns Unexpected Events

Spontaneous Services (P2P) Flash Crowds Network Attacks

Equipment Failures External Routing Changes

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 15

EuroFGI Workshop on IP QoS and Traffic Control

Examples of Variations in Real Data (I)

1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 Time (min) Link Load (unknown unit) Link 47 Link 58 Link 104 Link 126

Correlated volume changes Unidentifiable variations

(a) Traffic patterns in a large Tier-2 network.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 16

EuroFGI Workshop on IP QoS and Traffic Control

Examples of Variations in Real Data (II)

1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 0.5 1 1.5 2 2.5 x 10

3

Time (min) Link Load (Mbps)

External routing changes

(b) Traffic patterns in the Abilene network.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 17

EuroFGI Workshop on IP QoS and Traffic Control

Large heterogeneity in current traffic demands

200 400 600 800 1000 200 400 600 800 1000 1200 200 400 600 800 1000 3000 4000 5000 6000 7000 8000 9000 200 400 600 800 1000 500 1000 1500 2000 2500 200 400 600 800 1000 200 400 600 800 1000 1200 200 400 600 800 1000 1000 2000 3000 4000 5000 6000 7000 200 400 600 800 1000 100 200 300 400 500 600 700 200 400 600 800 1000 200 400 600 800 1000 1200 200 400 600 800 1000 100 200 300 400 500 600 200 400 600 800 1000 1400 1600 1800 2000 2200 2400 2600

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 18

EuroFGI Workshop on IP QoS and Traffic Control

The traditional approach for TE

Stable Prediction-Based Routing - Static Routing Based on Traffic Matrix (TM) estimation and prediction Relies on single estimated TM or group of expected TMs to

ptimize routing

Adaptive Routing - Dynamic Routing Load balancing in real-time Responds to instantaneous traffic demands, based on measurements.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 19

EuroFGI Workshop on IP QoS and Traffic Control

Multipath routing optimization

Consider the following network scenario : Network topology :

n nodes. L = {1, . . ., r} links with capacities in C = (c1, c2, . . . , cr). N = {OD1, .., ODm=n(n−1)} Origin-Destination traffic flows. Routing matrix R = {rl,k→ l=1..r,k=1..m}, 0 rl,k 1. P(k) = {set of paths p for ODk}, k = 1..m.

Traffic OD flows d = {di,j→ i,j=1..n} = ⇒ d = {dk, k=1..m} Links traffic (aggregated ODs traffic) y = {yl, l=1..r} y(t) = R × d(t) ∀t.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 20

EuroFGI Workshop on IP QoS and Traffic Control

Multipath routing optimization

Problem formulation

Given d, C, R and P(k), TE seeks to optimally balance d in P(k) to minimize some performance criterion :

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 21

EuroFGI Workshop on IP QoS and Traffic Control

Multipath routing optimization

Problem formulation

Given d, C, R and P(k), TE seeks to optimally balance d in P(k) to minimize some performance criterion : umax (C, d, R) = max

l∈{1...r}

k

rl,k · dk cl = max

l∈{1...r}

yl cl

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 22

EuroFGI Workshop on IP QoS and Traffic Control

Multipath routing optimization

Problem formulation

Given d, C, R and P(k), TE seeks to optimally balance d in P(k) to minimize some performance criterion : umax (C, d, R) = max

l∈{1...r}

k

rl,k · dk cl = max

l∈{1...r}

yl cl

xk

p , 0 xk p 1, fraction of

dk in p ∈ P(k) xk

l , 0 xk l 1, fraction of

dk in l ∈ p

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 23

EuroFGI Workshop on IP QoS and Traffic Control

Multipath routing optimization

Problem formulation

Given d, C, R and P(k), TE seeks to optimally balance d in P(k) to minimize some performance criterion : umax (C, d, R) = max

l∈{1...r}

k

rl,k · dk cl = max

l∈{1...r}

yl cl

xk

p , 0 xk p 1, fraction of

dk in p ∈ P(k) xk

l , 0 xk l 1, fraction of

dk in l ∈ p minimize umax subject to : P

p∈P(k)

xk

p

1 ∀ k ∈ N P

p∈P(k), l∈p

xk

p

xk

l

∀ k ∈ N, ∀ l ∈ L P

k∈N

xk

l .dk

umax · cl ∀ l ∈ L xk

p , xk l

∀ l ∈ L, ∀ p ∈ P(k), ∀ k ∈ N umax 1

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 24

EuroFGI Workshop on IP QoS and Traffic Control

The traditional approach for TE

Stable Prediction-Based Routing

ptimize routing for a single estimated TM → use it for

long-time periods [RTZ-03] traffic uncertainty is characterized by small sets of TMs (e.g. TMs from previous day, same day of previous week, etc.) [ZLGKMT-05]

[RTZ-03] M. Roughan, M. Thorup and Y. Zhang, “Traffic Engineering with Estimated Traffic Matrices”, IMC 2003. [ZLGKMT-05] C. Zhang, Y. Liu, W. Gong, J. Kurose, R. Moll and D. Towsley, “On Optimal Routing with Multiple Traffic Matrices”, INFOCOM 2005.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 25

EuroFGI Workshop on IP QoS and Traffic Control

The traditional approach for TE

Stable Prediction-Based Routing

ptimize routing for a single estimated TM → use it for

long-time periods [RTZ-03] traffic uncertainty is characterized by small sets of TMs (e.g. TMs from previous day, same day of previous week, etc.) [ZLGKMT-05] Remark : too optimistic approach for present dynamic demands.

[RTZ-03] M. Roughan, M. Thorup and Y. Zhang, “Traffic Engineering with Estimated Traffic Matrices”, IMC 2003. [ZLGKMT-05] C. Zhang, Y. Liu, W. Gong, J. Kurose, R. Moll and D. Towsley, “On Optimal Routing with Multiple Traffic Matrices”, INFOCOM 2005.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 26

EuroFGI Workshop on IP QoS and Traffic Control

The traditional approach for TE

Adaptive Routing

minimize network congestion,delays, etc by adaptively balancing the load among paths. desirable property → routing adapted to traffic.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 27

EuroFGI Workshop on IP QoS and Traffic Control

The traditional approach for TE

Adaptive Routing

minimize network congestion,delays, etc by adaptively balancing the load among paths. desirable property → routing adapted to traffic. Remark : difficult to implement, present poor performance under significant and abrupt traffic changes [WXQYZG-03].

[WXQYZG-03] H. Wang, H. Xie, L. Qiu, Y. Yang, Y. Zhang and A. Greenberg, “COPE : Traffic Engineering in Dynamic Networks”, SIGCOMM 2006.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 28

EuroFGI Workshop on IP QoS and Traffic Control

Outline

1

Traffic Engineering (TE) under traffic uncertainty

2

Robust and proactive TE : the Stable Robust Routing

3

A time-varying approach : the Multi-Temporal Robust Routing

4

Conclusions and Perspectives

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 29

EuroFGI Workshop on IP QoS and Traffic Control

A Proactive Approach

The Boy Scout motto :

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 30

EuroFGI Workshop on IP QoS and Traffic Control

A Proactive Approach

The Boy Scout motto :

Be Prepared!!!

Stable Robust Routing robust Traffic Engineering techniques. consider traffic uncertainty in advance (robustness). no routing changes (stability).

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 31

EuroFGI Workshop on IP QoS and Traffic Control

The Stable Robust Routing

Traffic uncertainty set and routing optimization

Traffic d is uncertain → belongs to an uncertainty set D : D = {d ∈ Rm, R × d = y, d 0}

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 32

EuroFGI Workshop on IP QoS and Traffic Control

The Stable Robust Routing

Traffic uncertainty set and routing optimization

Traffic d is uncertain → belongs to an uncertainty set D : D = {d ∈ Rm, R × d C, d 0}

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 33

EuroFGI Workshop on IP QoS and Traffic Control

The Stable Robust Routing

Traffic uncertainty set and routing optimization

Traffic d is uncertain → belongs to an uncertainty set D : D = {d ∈ Rm, R × d ymax, d 0}

minimize umax subject to : P

p∈P(k)

xk

p

1 ∀ k ∈ N P

p∈P(k), l∈p

xk

p

xk

l

∀ k ∈ N, ∀ l ∈ L P

k∈N

xk

l .dk

umax · cl ∀ l ∈ L, ∀ d ∈ D xk

p , xk l

∀ l ∈ L, ∀ p ∈ P(k), ∀ k ∈ N umax 1

[BAK-05] W. Ben-Ameur and H. Kerivin, “Routing of Uncertain Traffic Demands”, Optimization and Engineering, 2005.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 34

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization based on TME

At time to ⇒ real TM d∗, routing Ro and link loads yo D = {d ∈ Rm, Ro × d = yo, d 0}

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 35

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization based on TME

At time to ⇒ real TM d∗, routing Ro and link loads yo D = {d ∈ Rm, Ro × d = yo, d 0}

1. Ideal scenario

routing optimization for d∗ → u∗

max

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 36

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization based on TME

At time to ⇒ real TM d∗, routing Ro and link loads yo D = {d ∈ Rm, Ro × d = yo, d 0}

1. Ideal scenario

routing optimization for d∗ → u∗

max

2. Traditional TM estimation-based scenario

Ro and yo → estimated TM ˆ d (tomogravity) routing optimization for ˆ d → ˆ R ˆ umax = umax(C, d∗, ˆ R) ˆ uwc

max = max d∈D umax

C, d, ˆ

R

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 37

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization based on TME

At time to ⇒ real TM d∗, routing Ro and link loads yo D = {d ∈ Rm, Ro × d = yo, d 0}

1. Ideal scenario

routing optimization for d∗ → u∗

max

2. Traditional TM estimation-based scenario

Ro and yo → estimated TM ˆ d (tomogravity) routing optimization for ˆ d → ˆ R ˆ umax = umax(C, d∗, ˆ R) ˆ uwc

max = max d∈D umax

C, d, ˆ

R

3. Stable robust scenario

D → urobust wc

max

, Rrobust urobust

max

= umax (C, d∗, Rrobust)

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 38

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization based on TME

Day Time 02 :00 08 :00 14 :00 20 :00 ˆ umax 1.18 1.03 1.07 1.07 urobust

max

1.07 1.14 1.15 1.13 ˆ uwc

max

4.71 4.87 5.75 5.01 urobust wc

max

1.10 1.15 1.16 1.14 Routing performance, relative to u∗

max

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 39

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization based on TME

Day Time 02 :00 08 :00 14 :00 20 :00 ˆ umax 1.18 1.03 1.07 1.07 urobust

max

1.07 1.14 1.15 1.13 ˆ uwc

max

4.71 4.87 5.75 5.01 urobust wc

max

1.10 1.15 1.16 1.14 Routing performance, relative to u∗

max

Remark : are you ready to pay the price ?

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 40

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization based on TME

Day Time 02 :00 08 :00 14 :00 20 :00 ˆ umax 1.18 1.03 1.07 1.07 urobust

max

1.07 1.14 1.15 1.13 ˆ uwc

max

4.71 4.87 5.75 5.01 urobust wc

max

1.10 1.15 1.16 1.14 Routing performance, relative to u∗

max

Remark : are you ready to pay the price ? SRR offers worst-case guarantees at a reasonable cost (8%)

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 41

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization with time-varying TM

D = {d ∈ Rm, R × d ysafe, d 0} ysafe = yday

max

14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 0.1 0.2 0.3 0.4 0.5 0.6 Time (hours) Maximum Link Utilization Traditional Routing Stable Robust Routing Optimal Routing Traditional TE

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 42

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization with time-varying TM

D = {d ∈ Rm, R × d ysafe, d 0} ysafe = yday

max

14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 0.1 0.2 0.3 0.4 0.5 0.6 Time (hours) Maximum Link Utilization Traditional Routing Stable Robust Routing Optimal Routing Traditional TE

PBR → 60% degradation SRR → 11% degradation

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 43

EuroFGI Workshop on IP QoS and Traffic Control

Traditional vs. SRR in the Abilene Network

Routing optimization with time-varying TM

D = {d ∈ Rm, R × d ysafe, d 0} ysafe = yday

max

14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 0.1 0.2 0.3 0.4 0.5 0.6 Time (hours) Maximum Link Utilization Traditional Routing Stable Robust Routing Optimal Routing Traditional TE

PBR → 60% degradation SRR → 11% degradation

Remark : traditional TE is no longer suitable for current variable traffic.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 44

EuroFGI Workshop on IP QoS and Traffic Control

Trade off in the size of D

The size of D plays a key role in the cost (performance) of the SRR larger sets can handle broader groups of traffic demands, but at the cost of routing inefficiency. tighter sets produce more efficient routing schemes, but subject to poor performance guarantees.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 45

EuroFGI Workshop on IP QoS and Traffic Control

Trade off in the size of D

DA =

d ∈ Rm, R × d y4:00−18:00

max

, d 0

DB

=

d ∈ Rm, R × d y18:00−4:00

max

, d 0

5:00 7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00 1:00 3:00

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Time (hours) Maximum Link Utilization Historical Routing Robust Routing B Robust Routing A

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 46

EuroFGI Workshop on IP QoS and Traffic Control

Trade off in the size of D

DA =

d ∈ Rm, R × d y4:00−18:00

max

, d 0

DB

=

d ∈ Rm, R × d y18:00−4:00

max

, d 0

5:00 7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00 1:00 3:00

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Time (hours) Maximum Link Utilization Historical Routing Robust Routing B Robust Routing A

Remark : a single stable routing scheme for long-time periods results in sub-optimal performance.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 47

EuroFGI Workshop on IP QoS and Traffic Control

Outline

1

Traffic Engineering (TE) under traffic uncertainty

2

Robust and proactive TE : the Stable Robust Routing

3

A time-varying approach : the Multi-Temporal Robust Routing

4

Conclusions and Perspectives

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 48

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

The daily uncertainty set

time 00 :00 12 :00 24 :00 Dt

At each time t, R and y(t) = yt define an instantaneous uncertainty set :

D(t) =

d ∈ Rm, R × d yt, d 0
The continuous union of infinite instantaneous uncertainty sets along

time t defines the daily uncertainty set :

Dt =

(d, t) ∈ Rm+1, d ∈ Ut1ttτ D(t), t1 t tτ
P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 49

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Idea : divide the uncertainty set to reduce cost (adapt the set) and consider a SRR configuration for each sub-set [BA-07].

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 50

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Idea : divide the uncertainty set to reduce cost (adapt the set) and consider a SRR configuration for each sub-set [BA-07]. Partitioning hyperplane α.d = β.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 51

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Idea : divide the uncertainty set to reduce cost (adapt the set) and consider a SRR configuration for each sub-set [BA-07]. Partitioning hyperplane α.d = β. The optimal division is generally NP-complex.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 52

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Idea : divide the uncertainty set to reduce cost (adapt the set) and consider a SRR configuration for each sub-set [BA-07]. Partitioning hyperplane α.d = β. The optimal division is generally NP-complex. However, when the direction (α) is known, it can be approximately solved.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 53

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Idea : divide the uncertainty set to reduce cost (adapt the set) and consider a SRR configuration for each sub-set [BA-07]. Partitioning hyperplane α.d = β. The optimal division is generally NP-complex. However, when the direction (α) is known, it can be approximately solved. We take the time direction for the division : β = t.

[BA-07] W. Ben-Ameur, “Between Fully Dynamic Routing and Robust Stable Routing”, DRCN 2007.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 54

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set

γ + 1 hyper planes at times {β1, β2, .., βγ+1}

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 55

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set

γ + 1 hyper planes at times {β1, β2, .., βγ+1} Di = {Dt ∩ {d, α.d βi} ∩ {d, α.d βi+1}} , ∀i = 1, .., γ

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 56

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set

γ + 1 hyper planes at times {β1, β2, .., βγ+1} Di = {Dt ∩ {d, α.d βi} ∩ {d, α.d βi+1}} , ∀i = 1, .., γ D

′

i is the smallest set that contains Di, βi t βi+1

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 57

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set

γ + 1 hyper planes at times {β1, β2, .., βγ+1} Di = {Dt ∩ {d, α.d βi} ∩ {d, α.d βi+1}} , ∀i = 1, .., γ D

′

i is the smallest set that contains Di, βi t βi+1

ptimal routing changes β∗ =
β∗

2, .., β∗ γ

are the solution for :

time β1 β2 β3 D

′ 1

D

′ 2

β∗(Dt) = arg min

β

max

i=1..γ umax(D

′

i )

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 58

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set

γ + 1 hyper planes at times {β1, β2, .., βγ+1} Di = {Dt ∩ {d, α.d βi} ∩ {d, α.d βi+1}} , ∀i = 1, .., γ D

′

i is the smallest set that contains Di, βi t βi+1

ptimal routing changes β∗ =
β∗

2, .., β∗ γ

are the solution for :

time β1 β2 β3 D

′ 1

D

′ 2

β∗(Dt) = arg min

β

max

i=1..γ umax(D

′

i )

Generalized dichotomy method
P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 59

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set - an example

t_{min} t_{max} t_{med}

take tmin and tmax

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 60

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set - an example

t_{med}

tmin tmax take tmin and tmax βmin = tmin, βmax = tmax β = 1

2 (βmin + βmax)

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 61

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set - an example

tmin β tmax take tmin and tmax βmin = tmin, βmax = tmax β = 1

2 (βmin + βmax)

take D1 and D2

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 62

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set - an example

tmin β tmax take tmin and tmax βmin = tmin, βmax = tmax β = 1

2 (βmin + βmax)

take D1 and D2 u1(2)

max = urobust max

(D

′

1(2))

if u1

max > u2 max

βmax = β β = 1

2 (βmin + βmax)

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 63

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set - an example

tmin β tmax take tmin and tmax βmin = tmin, βmax = tmax β = 1

2 (βmin + βmax)

take D1 and D2 u1(2)

max = urobust max

(D

′

1(2))

if u1

max > u2 max

βmax = β β = 1

2 (βmin + βmax)

repeat until βmin ≈ βmax

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 64

EuroFGI Workshop on IP QoS and Traffic Control

The Multi-Temporal Robust Routing (MTRR)

Optimal division of the uncertainty set

Observation : the algorithm converges because : urobust

max

(Dt ∩ {d, α.d tmin} ∩ {d, α.d β}) is a non-decreasing function of β urobust

max

(Dt ∩ {d, α.d β} ∩ {d, α.d tmax}) is a non-increasing function of β

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 65

EuroFGI Workshop on IP QoS and Traffic Control

Some results in the Abilene Network

21:00 23:00 1:00 3:00 5:00 7:00 9:00 11:00 13:00 17:00 19:00 21:00 0.2 0.25 0.3 0.35 0.4 0.45 Time (hours) Maximum Link Utilization Historical Routing Stable Robust Routing A Stable Robust Routing B M−T Robust Routing

(a) Expected daily behavior

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 66

EuroFGI Workshop on IP QoS and Traffic Control

Some results in the Abilene Network

21:00 23:00 1:00 3:00 5:00 7:00 9:00 11:00 13:00 17:00 19:00 21:00 0.2 0.25 0.3 0.35 0.4 0.45 Time (hours) Maximum Link Utilization Historical Routing Stable Robust Routing A Stable Robust Routing B M−T Robust Routing

(a) Expected daily behavior

β∗ ≈ 8 : 00 MTRR improvements :

16% before β∗ 20% after β∗

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 67

EuroFGI Workshop on IP QoS and Traffic Control

Some results in the Abilene Network

5:00 7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00 1:00 3:00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Time (hours) Maximum Link Utilization Historical Routing Stable Robust Routing A Stable Robust Routing B M−T Robust Routing

(b) Unexpected and large variation

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 68

EuroFGI Workshop on IP QoS and Traffic Control

Some results in the Abilene Network

5:00 7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00 1:00 3:00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Time (hours) Maximum Link Utilization Historical Routing Stable Robust Routing A Stable Robust Routing B M−T Robust Routing

(b) Unexpected and large variation

β∗ ≈ 18 : 00 MTRR improvements reach almost 60%

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 69

EuroFGI Workshop on IP QoS and Traffic Control

Outline

1

Traffic Engineering (TE) under traffic uncertainty

2

Robust and proactive TE : the Stable Robust Routing

3

A time-varying approach : the Multi-Temporal Robust Routing

4

Conclusions and Perspectives

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 70

EuroFGI Workshop on IP QoS and Traffic Control

Conclusions of this work

Traditional TE mechanisms for routing optimization may no longer be applicable for current traffic scenario. Stable Robust Routing can be used for routing optimization under traffic uncertainty, but it is not enough. The Multi-Temporal Robust Routing outperforms the SRR approach, but it is necessary to have a rough knowledge of traffic variations to apply it (daily uncertainty set). Reactive methods should be applied to manage unexpected events, but smartly combined with robust routing techniques to avoid instabilities.

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 71

EuroFGI Workshop on IP QoS and Traffic Control

Some future directions

study the impacts of the MTRR in QoS (losses, delay, etc.). combine the MTRR (routing optimization for normal traffic behaviors, but with guarantees) with reactive routing techniques (detection of large abnormal events and accurate routing reconfiguration).

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE

SLIDE 72

EuroFGI Workshop on IP QoS and Traffic Control

Thank You for Your Attention ! ! Remarks & Questions ?

P. CASAS - S. VATON

Computer Science Department ENST BRETAGNE