istituto di scienza e tecnologie dellinformazione a faedo september 26th 200713 th ka-band...

September 26th 2007 13th Ka-band conference Istituto di Scienza e Tecnologie dell’Informazione “A Faedo”

TOWARD A COMPLETE CONTROL FRAMEWORK FOR ADAPTIVE CAPACITY

ALLOCATION IN MOBILE DVB-RCS

Nedo Celandroni*, Franco Davoli°, Erina Ferro*, Alberto Gotta*, Raffaello Secchi*

*ISTI-CNR, Area della Ricerca del C.N.R., Via Moruzzi 1, I-56124 Pisa, Italy

°DIST-University of Genoa and CNIT (Italian National Consortium for Telecommunications), Italy


Objectives

• We address the general problem of controlling and optimizing access and bandwidth sharing among different applications in mobile DVB-RCS

• This work deals explicitly with the dynamic assignment of bandwidth to elastic (TCP) traffic on the uplink. Different access techniques are compared:– Random Access: Diversity Slotted Aloha (DSA)– Deterministic Access: Demand Assignment Multiple Access (DAMA)

• The convenience of one technique with respect to the other as concerns TCP performance is determined by:– Bandwidth Cost per connection

• Total number of slots allocated in the case of DAMA• Data successfully received in DSA case divided by DSA throughput

– Average Completion Time


MUltimedia MObile Bandwidth ALlocation (MUMOBAL) architecture

• Class-based Dynamic Bandwidth Allocation– One portion of bandwidth for each typology of traffic and moving boundaries between

bandwidth portions• In each portion, the Network Control Centre (NCC) assigns bandwidth to the relevant

traffic classes, by taking into account their QoS requirements and traffic characteristics

Collision free Optimal Controller

Collision freeDAMA

Contention accessDSA

Collision free Optimal Controller

Collision freeDAMA

Contention accessDSA

•EF traffic:• Voice, video & Real Time

applications with QoS requirements•Large volume AF & BE traffic:

• web applications, video streaming, interactive and peer-to-peer, etc...

•Small volume BE traffic:• web-browsing, e-mail, signaling, etc…

•EF traffic:• Voice, video & Real Time

applications with QoS requirements•Large volume AF & BE traffic:

• web applications, video streaming, interactive and peer-to-peer, etc...

•Small volume BE traffic:• web-browsing, e-mail, signaling, etc…

•Collision free Optimal controller:• Optimal DBA Control Algorithm

for Service Level Agreement •Collision free DAMA:

• Rate Based Dynamic Capacity• Volume Based Dynamic Capacity• FCA option

•Contention access DSA:• Contention-based MAC

•Collision free Optimal controller:• Optimal DBA Control Algorithm

for Service Level Agreement •Collision free DAMA:

• Rate Based Dynamic Capacity• Volume Based Dynamic Capacity• FCA option

•Contention access DSA:• Contention-based MAC


DAMA scheme

• DAMA scheme adopted resembles the FIFO Ordered Demand Assignment (FODA)

– N. Celandroni, E. Ferro, “The FODA-TDMA satellite access scheme: presentation, study of the system and results”, IEEE Transactions on Communications, December 1991

• The bandwidth requests, contained in each packet header, is done by each station taking into account the instantaneous queue size and a term proportional to the incoming traffic

• The NCC performs the bandwidth assignment by cyclically scanning all requests and giving each station an amount of bandwidth proportional to the request

– The assigned amount of bandwidth is subtracted from the request for the next assignment cycle; successive requests coming from the stations override the current residual values of the previous ones

– The length of the cycle is dynamically varied according to the overall system load – A lower bound on the assignment is set as a minimum guaranteed bandwidth,

whereas an upper bound is given by the capacity of a carrier


Diversity Slotted Aloha (1/2)

timei-th frame

station #1 station #2 station #3

time slot

DSA for k=2A packet is sent twice


Diversity Slotted Aloha (2/2)

pc(k ) 1 e kG k

(k ) G 1 pc(k )

• The NCC maintains a target throughput by allocating the number of DSA slots per frame that are necessary to achieve it; it corresponds to the average number of successful slots in a frame divided by the target throughput itself

• For design purposes, we can consider the asymptotic values of DSA collision probability and throughput as functions of G, the number of users per slot, and k, the number of copies delivered


Case study: mobile user acts as an FTP client

• Assumptions– Users are synchronized with the system and receive the burst time plan, sent

at each frame time by the NCC in TDM mode (DVB-S2)– Any user is allowed to transmit in a frame only upon receiving a valid burst

time plan for that frame

• DAMA first access & connection termination– A user issues the initial request for bandwidth on reception of a SYN

message from the TCP server by sending a request packet in the DSA area– Successive requests for bandwidth are piggy-backed with data – A user releases the bandwidth when the data transfer is completed

• System parameters– DAMA: the minimum amount of bandwidth equal to one slot every m frames

allocated by the NCC upon the reception of the access request – DSA: the target throughput and, consequently, the collision probability that

derives by the best choice of the k value• Target throughput 10% k=4, collision probability = 0.012• Target throughput 20% k=3, collision probability = 0.121


Simulative Scenario (ns-2)

HUB station

Internet Gateway

Terrestrial

Segment (50 ms)

Environment TNB TBLBlocking

prob.

Highway 3.02 s 0.36 s 0.107

Rural 2.03 s 0.55 s 0.215

CLIENT

LMS channel parameters

SERVER

UPLINK: TDMA/DSADOWNLINK: TDM

ON/OFFmobile channel

RTT=600 ms, Segment size=1416B (equivalent to 8 DVB packets), Frame size=20 ms, TCP/Sack


Short-lived connections (highway)

2

4

6

8

10

12

14

0 20 40 60 80

Connection Length [kB]

Com

plet

ion

Tim

e [s

]

DSA 10%DSA 20%TDMA 1f/sTDMA 2f/sTDMA 4f/sTDMA 8f/s

0

100

200

300

400

500

600

700

0 20 40 60 80


DV

B p

acke

ts/c

onne

ctio

n

DSA 10%DSA 20%TDMA 1f/sTDMA 2f/s

TDMA 4f/sTDMA 8f/s

Average Uplink Bandwidth Cost Average Completion Time


Short-lived connections (rural)

0

100

200

300

400

500

600

700

0 20 40 60 80


DV

B p

acke

ts/c

onne

ctio

n


TDMA 4f/sTDMA 8f/s

2

7

12

17

22

0 20 40 60 80


Com

plet

ion

Tim

e [s

]


TDMA 4f/sTDMA 8f/s

Average Uplink Bandwidth Cost Average Completion Time


Long-lived connections (highway)

Completion Time

0

200

400

600

800

1000

1200

1400

1600

10% 20% 1 f/s 2 f/s 4 f/s 8 f/s

dura

tion

[s]

• DSA may not be suitable for long-lived connections in highway– DSA 10% is outperformed by DAMA 1f/s in both duration and cost, while

DSA 20% has a duration very close to DAMA 4f/s, but the latter exhibits a significantly lower cost

Bandwidth Cost

0

500010000

15000

2000025000

30000

3500040000

45000

10% 20% 1 f/s 2 f/s 4 f/s 8 f/s

DV

B p

acke

ts/c

onne

ctio

n

DSA TDMA

DSA TDMA


Long-lived connections (rural)

• In a rural environment – DSA 10% and DAMA 2f/s exhibit very similar performance in both

duration and cost, while DSA 20% has a duration similar to DAMA 8f/s, but the latter exhibits a significantly lower cost

Bandwidth Cost

0

10000

20000

30000

40000

50000

60000

70000

80000

90000

10% 20% 1 f/s 2 f/s 4 f/s 8 f/s

DV

B p

acke

ts/c

onne

ctio

n

DSA TDMA

Completion Time

0

500

1000

1500

2000

2500

3000

10% 20% 1 f/s 2 f/s 4 f/s 8 f/s

dura

tion

[s]

DSATDMA


Summary of simulations results

1. The smaller the size of TCP messages, the higher the gain of DSA with respect to DAMA in bandwidth expenditure

2. Since DSA cost depends on data successfully delivered only, it is less sensible to changes in the channel blocking probability

3. DSA is convenient for very small transfers in terms of completion time in both environments as well– the connection duration in DAMA 1f/s is comparable with DSA 10% in that

very low losses of ACKs due to collision cause negligible performance degradations to TCP

4. Limited to our investigations, DSA is not convenient for long-lived connections in both environments

5. The choice of system parameters is a matter of more complex investigations (e.g. the mobile acts as server). Nevertheless, once the parameters have been selected, cost diagrams will be used to evaluate convenience thresholds. For instance– In case DAMA 4f/s and DSA10% are selected as the most convenient

options, the threshold file size is about 17 kB and 65 kB for highway and rural environments, respectively


Conclusions & Future works

• We analyzed the behaviour of two different access methods, based on DSA and DAMA respectively, in providing the basis for a FTP

– The two methods have been compared in terms of bandwidth cost and connection completion time under the case of users acting as clients

– Simulation results have shown an advantage of DSA over DAMA up to certain convenience thresholds in the file size as functions of DSA throughput and DAMA m parameter

– The non compliance with the convenience thresholds would imply an increased cost for users and, likely, higher revenue for the service provider

– DSA does not result convenient for long-lived connections in both environments

• This study will be completed with the analysis of the case in which the mobile user acts as a server

– Server and client cases are necessary to operate the choice of the target throughput in DSA and the m factor in DAMA

– For the server case, in order to obviate to the collisions’ effect in DSA, a packet level FEC might be necessary

istituto di scienza e tecnologie dellinformazione a faedo september 26th 200713 th ka-band...

Documents

bandwidth assignment

bandwidth requests

frame dama

portion of bandwidth

bandwidth proportional

dynamic bandwidth allocation

frame time

bandwidth cost