08.09.2011 The Green-Game: Striking a Balance between QoS ... · Striking a Balance between QoS and Energy Saving ... our work focusses on “energy aware ... -Automatic adaptation

The GreenStriking a Balance between QoS and

Energy SavingEnergy Saving

Aruna Prem Bianzino1, Claude Chodet

The Green-Game: Striking a Balance between QoS and

Energy Saving

08.09.2011

Energy Saving

, Claude Chodet1, Stefano Moretti2, Dario Rossi1, Jean-Louis Rougier1

1. Telecom ParisTech, 2. Université Paris-Dauphine

Introduction: ICT for Green

• Information and Communication Technologies (ICT) offer big opportunities to reduce the human footprint:

- Videoconferences, telepresence, email, etc.

The 23rd International Teletraffic Congress - ITC 2011 - San Francisco, USA

Introduction: ICT for Green

Information and Communication Technologies (ICT) offer big opportunities to reduce the human footprint:

Videoconferences, telepresence, email, etc.

08.09.2011

San Francisco, USA 2/18

Introduction: Green for ICT

• ICT itself represents a strong contributionenvironmental impact of human activities, and with a very high increasing rate:

- Same footprint of the airplane transports,

… but with higher growing rate.… but with higher growing rate.

Remark: our work focusses on “energy aware

Gas emission is complex to quantify (type of energy, …)

Economical arguments (reduce energy cost)


Introduction: Green for ICT

strong contribution to the environmental impact of human activities, and with a very

Same footprint of the airplane transports,

… but with higher growing rate.

08.09.2011

… but with higher growing rate.

energy aware” ICT

Gas emission is complex to quantify (type of energy, …)

Economical arguments (reduce energy cost)


Green for ICT: A Hot Topic

Many works have been initiated in the last years:In Data Centers, in Peripherals

• In wireless networks, not completely a subject:

- Battery constraints in wireless mesh/sensor networks- Interferences (power control)- Important savings

• In wired networks:- Still some interesting opportunities - Depend on topologies


Green for ICT: A Hot Topic

Many works have been initiated in the last years:Peripherals, and in Networks:

, not completely a “new”

Battery constraints in wireless mesh/sensor

08.09.2011

Interferences (power control)

Still some interesting opportunities Depend on topologies


Energy Saving Opportunities

Facts:- Network systems and devices are - Predicable traffic fluctuations

- Today: energy agnostic equipments- How to reach proportionality (energy/utilization)

Utilisation

Pow

er c

onsu

mpt

ion



Network systems and devices are over-provisionedfluctuations

08.09.2011

Today: energy agnostic equipmentsHow to reach proportionality (energy/utilization)

WorstIntermediateIdeal


Resource Consolidation

- Already a popular practice in other domains(e.g., Data Centers)

Traffic

- Traffic aggregation through a proper

Time

Traffic


Resource Consolidation

Already a popular practice in other domains

08.09.2011

Traffic aggregation through a proper weight optimization.


Resource Consolidation: Open Points

- A solution purely optimizing the energy consumption does not take into account the

- There is no control on which network elements are switched off

- Definition of a criticality index - Definition of a criticality index devices to drive the resource consolidation process

Definition of a trade off between

Energy-saving


Resource Consolidation: Open Points

A solution purely optimizing the energy consumption does not take into account the system robustness

There is no control on which network elements are

criticality index for the network

08.09.2011

criticality index for the network devices to drive the resource consolidation process

Definition of a trade off between

saving and robustness


A Game-Theoretical Approach

- The basic idea:

• Modeling the communication network as a cooperative TU-game

• Each node is a player

• Every coalition is a network configuration:

- Nodes in the coalition

- Other nodes -> OFF (or failures)

• The amount of delivered traffic is the revenue of the coalition


Theoretical Approach

Modeling the communication network as a

08.09.2011

Every coalition is a network configuration:

Nodes in the coalition -> ON

OFF (or failures)

The amount of delivered traffic is the revenue of


A Game-Theoretical Approach (2)

- The final game is the composition of two games:

• A Traffic Game (A-Priori) over a full(allows all coalitions, accounts only for the Traffic Requests)

• A Topology Game (A-Posteriori), which is the • A Topology Game (A-Posteriori), which is the restriction of the first over the network graph, and accounts for the Topology

• The two games may be decomposed into several unanimity games


Theoretical Approach (2)

The final game is the composition of two games:

Priori) over a full-mash network (allows all coalitions, accounts only for the Traffic

Posteriori), which is the

08.09.2011

Posteriori), which is the restriction of the first over the network graph, and accounts for the Topology

The two games may be decomposed into several


The Shapley Ranking

- The Shapley value defines a rank among players (on the basis of the amount of traffic that nodes contribute to carry, and of their criticality while composing the coalition)

- Nodes are progressively switched off (if the all traffic - Nodes are progressively switched off (if the all traffic requests are still satisfied, with eventual maximum load constraints)

- When to stop?

CF A

Lower Shapley


The Shapley Ranking

defines a rank among players (on the basis of the amount of traffic that nodes contribute to carry, and of their criticality while composing the

Nodes are progressively switched off (if the all traffic

08.09.2011

Nodes are progressively switched off (if the all traffic requests are still satisfied, with eventual maximum load

B D E

Higher Shapley


The Shapley Ranking: Toy Cases

1)

2) D

B

A

3)

C

DA

B

CE


The Shapley Ranking: Toy Cases

Every node has the same Shapley value

A, D -> 5/12

C, B -> 1/12

08.09.2011

->

A, D -> 23/60

B -> 8/60

C, E -> 3/60


Other Possible Rankings

- Other criticality indexes are present in the literature, but all of them only account for the network topology

- The G-Game considering an uniform TM matches quite well these indexes, but there is low correlation when taking into account the Trafficwhen taking into account the Traffic

G-Game (U-TM) Betweenness Degree

G-Game (U-TM) 1

Betweenness 0.9688 1

Degree 0.4594 0.5321 1

Closeness 0.8729 0.9057 0.6216

Eigen -0.0073 0.0792 0.7335

G-Game 0.4085 0.4286 0.2527

Traffic Load 0.4251 0.4868 0.4762


Other Possible Rankings

Other criticality indexes are present in the literature, but all of them only account for the network topology

Game considering an uniform TM matches quite well these indexes, but there is low correlation when taking into account the Traffic

08.09.2011

when taking into account the TrafficCloseness Eigen G-Game Traffic

Load

1

0.1787 1

0.5132 -0.0220 1

0.6046 0.1911 0.5583 1


A Real Case Study: The Network Scenario

- TIGER2 Network (typical access/metro network)

- Access nodes (traffic Sources - Access nodes (traffic Sources and Destinations)

- Core nodes (only traffic transport)

i

j


A Real Case Study: The Network Scenario

TIGER2 Network (typical

nodes (traffic Sources

1 6

78

9

1 0 1 8

1 9 2 02 1

08.09.2011

nodes (traffic Sources

2

3

1 3

4

51 6

1 1

1 2

1 4 1 5

1 7T


A Real Case Study: Different Rankings

1 6

78

9

1 0

1 1 1 7

1 8

1 9 2 02 1

T

2

3

1 3

4

51 6

1 1

1 2

1 4 1 5

1 7T

Only Topology


A Real Case Study: Different Rankings

T1 1

1 8

1 91 6

78

9

1 0

1 7

2 02 1

08.09.2011

T1 1

1 2

2

3

1 3

4

51 6

1 4 1 5

1 7

G-Game


A Real Case Study: Results


A Real Case Study: Results S U L

9 5 9

15 9 15

13 14 8

14 20 7

16 21 5

21 15 4

11 13 21

10 11 20

20 10 2

19 4 3

17 16 1

08.09.2011

17 16 1

18 19 6

12 6 14

8 1 11

5 12 10

7 17 19

4 8 16

3 T 13

6 7 12

1 3 17

2 18 18

T 2 T


A Real Case Study: Results (2)

10

15

20

25

30

35

40

45

Per

cent

age

of li

nk

Order Shapley

Energy saving (%) 17.05

Weighted avg path length 2.99


0

5

10

[0-10)[10-20)[20-30)[30-40)[40-50)[50-60)

Per

cent

age

of li

nk

Link load (%)

A Real Case Study: Results (2) S U L

9 5 9

15 9 15

13 14 8

14 20 7

16 21 5

21 15 4

11 13 21

10 11 20

20 10 2

19 4 3

17 16 1

08.09.2011

U-TMLoad

ShapleyReference

17 16 1

18 19 6

12 6 14

8 1 11

5 12 10

7 17 19

4 8 16

3 T 13

6 7 12

1 3 17

2 18 18

T 2 T

U-TM Load

13.43 16.27

3.40 3.25


[50-60)[60-70)[70-80)[80-90)[90-100]

Link load (%)

G-Game: Future Work

- Evaluation of the impact of:

• different scenarios (topology and TM), to highlight dependences

• different energy models (e.g. optical transmission, with periodical signal regeneration)with periodical signal regeneration)

- Evaluation of the effects on the inter(hot potato)

- Integration with dynamic routing protocols (IGP Weight Optimization)


Game: Future Work

Evaluation of the impact of:

different scenarios (topology and TM), to highlight

different energy models (e.g. optical transmission, with periodical signal regeneration)

08.09.2011

with periodical signal regeneration)

Evaluation of the effects on the inter-domain routing

Integration with dynamic routing protocols (IGP


Thank you!

Further details at:perso.telecom

Thank you!

Further details at: [email protected]/~bianzino

The G-Game: An Example

1)

2)

D

C

B

A

B2)

3)

D

C

B

A

D

C

B

A


Game: An Example

Traffic Request

A -> D: 50Mbps

Traffic Game:

08.09.2011

Traffic Game:

v = u(A, D)

Topology Game:

Minimal connected components containing A, D:

vΓ = u(A,B,D)+u(A,C,D)-u(A,B,C,D)



Politecnico di Torino, Italy 29/06/2010


6/27Politecnico di Torino, Italy 29/06/2010

EAR: An Optimization Approach

-Modeling the device energy consumption as a function of the utilization level

- ILP formulation for a - ILP formulation for a minimum-energy routing


- Evaluation over different network scenarios (i.e., topologies and traffic matrixes)

- Accounting for QoS: maximum imposable link load

EAR: An Optimization Approach

the device energy consumption as a function of


Evaluation over different network scenarios (i.e., topologies and traffic matrixes)

Accounting for QoS: maximum imposable link load

EAR: Results

- The solution has been tested over the GEANT2 network with real traffic matrixes

- All nodes generate/receive traffic (unfriendly scenario)traffic (unfriendly scenario)


The solution has been tested over the GEANT2 network

All nodes generate/receive traffic (unfriendly scenario)traffic (unfriendly scenario)


EAR: Results (3)


• The work has been finalized in:

- A.P. Bianzino, C. Chaudet, F. Larrocca, D. Rossi, J.submitted to GreenComm3 (GLOBECOM), Florida, USA,

- A.P. Bianzino, C. Chaudet, D. Rossi, J.-L. Rougier, Charge Network Flow Formulation”, in ACM SIGCOMM e2010.

10

15

20

25

30

35

40

45

Per

cent

age

of li

nk

IGP-WOGreen (α=0.75)


0

5

10[0]

(0-10](10-20](20-30](30-40](40-50](50-60](60-70](70-80](80-90](90-100]

Per

cent

age

of li

nk

Link load (%)

A.P. Bianzino, C. Chaudet, F. Larrocca, D. Rossi, J.-L. Rougier, “Energy-Aware Routing: a Reality Check”, Florida, USA, Dec. 2010.

L. Rougier, “Energy-Awareness in Network Dimensioning: a Fixed ACM SIGCOMM e-Energy'10, Extended Abstract, Passau, Germany, April

EAR: What is next?

• Formulation:

- Use of approximate solutions

- Comparison with existing heuristics

• Refinement of the model:

- Evaluation of the robustness of the solution


- Evaluation of the robustness of the solution

- Considering optical transmission: periodical signal regeneration, thus length dependent energy consumption

• Result Analysis:- Impact of Topology: Considering a broader set of scenarios (topologies, and TMs)

- Evaluation of the effects on the inter

Use of approximate solutions

Comparison with existing heuristics

Refinement of the model:

Evaluation of the robustness of the solution


Evaluation of the robustness of the solution

Considering optical transmission: periodical signal regeneration, thus length dependent energy consumption

Impact of Topology: Considering a broader set of scenarios (topologies, and TMs)

Evaluation of the effects on the inter-domain routing

Other Interesting Solutions: ALR

- Automatic adaptation of the link rate to its real utilization

- Different possible algorithms, different possible states/rates

S

states/rates

- A comparison is missing

- Evaluation of the interactions with the TCP congestion control

- Evaluation of the effects on the QoS


Other Interesting Solutions: ALR

Automatic adaptation of the link rate to its real

Different possible algorithms, different possible

DL

Evaluation of the interactions with the TCP

Evaluation of the effects on the QoS


Other Interesting Solutions: Proxying

- AEx: ARP, BitTorent, …

- Allows terminals to enter into sleep mode


Other Interesting Solutions: Proxying

Allows terminals to enter into sleep mode

A1Politecnico di Torino, Italy 29/06/2010

EAR: Further Results (GEANT2)


EAR: Further Results (GEANT2)

A2Politecnico di Torino, Italy 29/06/2010

Documents

08.09.2011 The Green-Game: Striking a Balance between QoS ... · Striking a Balance between QoS and Energy Saving ... our work focusses on “energy aware ... -Automatic adaptation