Multi-terminal dynamic bandwidth allocation in ... - Semantic Scholar

Outline

Multi-terminal dynamic bandwidth allocation in GEO Satellite Networks L. Chisci1

R. Fantacci2

F. Francioli2

T. Pecorella3

1 Dpt.

di Sistemi e Informatica Universit` a di Firenze, Italy.

2 Dpt.

di Elettronica e Telecomunicazioni Universit` a di Firenze, Italy.

3 CNIT

- Unit` a di Firenze, Italy.

e-mail: [email protected], {fantacci,francioli,pecos}@lenst.det.unifi.it

The 59th IEEE Vehicular Technology Conference, Spring 2004 Chisci, Fantacci, Francioli, Pecorella

Multi-terminal dynamic bandwidth allocation in GEO Sat.

Outline

Outline 1

Satellite Systems

2

Quality of Service (QoS)

3

Dynamic Bandwidth Allocation (DBA) DBA Scheme Satellite Gateway (SG) Network Control Center (NCC)

4

Results

5

Conclusions

Chisci, Fantacci, Francioli, Pecorella


Satellite Systems Quality of Service (QoS) Dynamic Bandwidth Allocation (DBA) Results Conclusions

Satellite Networks Pros Easy deployment Cost-effective Mobility Broadcasting Geographical coverage Cons In GEO links propagation delay is very high (RTT ' 500 ms). High delay*bandwidth systems are critycal for TCP flows. The bandwidth is scarce, so a careful management is needed. Chisci, Fantacci, Francioli, Pecorella



Quality of Service (QoS) QoS metrics Bandwidth Delay Jitter

Application VoIP Streaming video Web E-mail

Required Band Low High Avg. Low

Sensitivity to Delay Jitter Loss High High Avg. Avg. Avg. Avg. Avg. Low High Low Low High

Packet Loss

Internet Engineering Task Force (IETF) proposals Integrated Services (IntServ) Differentiated Services (DiffServ)








Packet Loss









Packet Loss





Dynamic Bandwidth Allocation Scheme Satellite Gateway (SG) Network Control Center (NCC)

Dynamic Bandwidth Allocation (DBA) Scheme

NCC

delay/2

delay/2

Satellite Gateways (SGs) can be divided in different priority classes, e.g.

delay/2

delay/2

delay/2

delay/2

Gold Silver Bronze

SG1

SG2

SGN





Bandwidth request algorithm to the NCC

Satellite Gateway Adaptive Traffic Predictor

ˆ (t + k | t) w

Bandwidth Controller

u(t)

y(t) bi

w(t)

Traffic Predictor

- v1(t) FIFO Buffer

- v2 (t)

In order to compute the bandwidth requests, each SG use 2 blocks:

Scheduler

- vn (t)


v(t) from the NCC

Bandwidth Controller




Buffers (FIFO queues) Input packets are divided into n buffers (FIFO queues), according to the DiffServ policy. (e.g. EF, AF, BE classes =⇒ n = 3)

 y(t) ∈ IRn vector of queue sizes  w(t) ∈ IRn vector of input flows at sample time t  n v(t) ∈ IR vector of output flows

y(t + 1) = y(t) + w(t) − v(t)





Scheduler v (t): bandwidth assigned at time t to the SG from the NCC v(t) ∈ IRn : transmission flows at time t for the n service classes   b1 n  b2  X   bi ≥ 0, bi ≤ 1 v(t) =  .  v (t),  ..  i=1

bn Several scheduling policies can be adopted, e.g. WFQ - Weighted Fair Queueing Priority Scheduling Weighted Round Robin Chisci, Fantacci, Francioli, Pecorella




Traffic Predictor

ˆ + k|t): predictions, at time t, of the input traffic flows at time w(t 4

t + k, based on the available data wt−1 = {w(k), k ≤ t − 1}.

Several predictors can be used: ARMA predictor (not self-similar) FARIMA predictor (self-similar) based on a fractionally integrated ARMA model Maximum Likelihood predictor





Bandwidth Controller The bandwidth request u(t) for the time t + δ, where δ is the Round Trip Delay, is based on the data already known at time t and assuming that the request is always satisfied, i.e. v (t) = u(t − δ) 0 state: x(t) = y0 (t), u(t − 1), . . . , u(t − δ) ∈ IRn+δ Predictive model for time evolution of queue sizes: ( x(t + k + 1) = Ax(t + k) + Bu(t + k) + Ew(t + k) y(t + k) = Cx(t + k) ˆ + k|t) are used in place of w(t + k), k ≥ 0. Predictions w(t Chisci, Fantacci, Francioli, Pecorella




Bandwidth Controller (cont.) u(t) is chosen to trade-off small queue sizes (i.e. low traffic delays) vs. bandwidth waste =⇒ Cost minimization of J =

H−1 X

y0 (t + δ + k) Q y(t + δ + k) + r u 2 (t + k)

k=0

subject to the constraints: ( ymin ≤ y(t + δ + k) ≤ ymax umin ≤ u(t + k) ≤ umax


k = 0, 1, . . . , H − 1




Receding Horizon control strategy Receding Horizon control is used to compensate the model uncertainties. Being u(t), . . . , u(t + H − 1) the optimal solution at time t we apply the request u(t) discarding u(t + 1), . . . at time t + 1 we repeat the process with the new data.





Other control strategies Predictive only Bandwidth request is based only on Predictor forecasting: ˆ + δ|t) u(t) = w(t Reactive controller Bandwidth request is computed using a Smith Predictor: " # t−1 X δ∗ w (k) u(t) = w (t) + K y (t) − u(k) − 1 − δ k=t−δ

where δ ∗ ≥ δ is a parameter representing the desirable queueing delay while K is a gain which, for stability, must belong to the interval (0, 1]. Chisci, Fantacci, Francioli, Pecorella




NCC bandwidth assignment The bandwidth is shared among SGs following a simple rule: 4

ui = ui (t − δ/2): bandwidth requested from SGi 4

vi = vi (t + δ/2): bandwidth assigned to SGi C : link capacity (max. bandwidth × sampling time T )

Constraints: PN i=1 vi ≤ C vi ≤ ui

i = 1, 2, . . . , N





Bandwidth assignment policy if

PN

ui ≤ C set vi = ui

if

PN

ui > C set vi = vi (u1 , u2 , . . . , uN )

i=1 i=1

function vi (u1 , u2 , . . . , uN ) depends on the assignment policy Weighted Proportional Assignment (WPA) Shares out the link capacity C proportionally to the requests with weights αi > 0 depending on the class of SGi (e.g. Gold, Silver and Bronze SGs). αi ui C vi = PN j=1 αj uj




Traffic delays, simulated congestion 0.25

Packet Delay [s]

0.2

5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 950

0.15

RHC Smith Pred

0.1

EF

0.05 0

1000

1050

1100

1150

1200

1250

1300

Simulation Time [s]

Faster recovery (1 vs. 6 min of Pred) Lower steady-state than Smith’s Chisci, Fantacci, Francioli, Pecorella

4 3.5 3 2.5 2 1.5 1 0.5 0

16 14 12 10 8 6 4 2 0

AF

BE



Bandwidth waste, simulated congestion 14

10 8

RHC Smith Pred

6 4 2 0 950

1000

1050

1100

1150

1200

1250

1300

Simulation Time [s]

Comparable band waste with Pred (' 0.5%) Smith’s band waste is ' 12% Chisci, Fantacci, Francioli, Pecorella

Packet Delay [s]

Band Lost [%]

12

5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 950

RHC Smith Pred

1000

1050

1100

1150

1200

1250

1300

Simulation Time [s]



Traffic delays, simulated congestion, SGs classes 7

Gold Silver Bronze

Packet Delay [s]

6 5

WPA weights:

4 3

αg = 1.0

2

αs = 0.8

1

αb = 0.6

0 900

950

1000

1050

1100

1150

1200

Simulation Time [s]

NCC algorithm can provide different QoS to each SG class.




Conclusions DBA for GEO satellite systems based on predictive control algorithm run in parallel in each Gateway bandwidth assignment algorithm in the NCC Interesting results in terms of multiple traffic classes management with different QoS low bandwidth waste high performance in steady state and after congestion Future developments different assignment policies in the NCC self-similar traffic predictors game-theoretic approach to DBA Chisci, Fantacci, Francioli, Pecorella






