Hasan SÖZER1 Data Scheduling and SAR for Bluetooth MAC Manish Kalia, Deepak Bansal, Rajeev Shorey IBM India Research Laboratory.

Hasan SÖZER 1

Data Scheduling and SARfor Bluetooth MAC

Manish Kalia, Deepak Bansal, Rajeev Shorey

IBM India Research Laboratory

Hasan Sözer 2

Outline

• Medium Access Control in Bluetooth

• Problems & Restrictions faced in Bluetooth MAC

• Goals, Assumptions & Approaches

• Priority Policy (PP)

• K-Fairness Policy (KFP)

• Scheduling Data in Presence of Voice

• Bluetooth SAR Policy & Possible Improvements

• Results & Conclusion

Hasan Sözer 3

Medium Access Control in Bluetooth

• TDD slot structure with strict alternation of slots between the Master and the Slaves

• Single point of coordination (at Master)

• Polling based

• A slave transmits packets in the reverse slot only after the Master polls the slave in a forward slot

• Thus, Bluetooth is a Master driven, polling based TDD standard

Hasan Sözer 4

Problems & Restrictions

• Conventional scheduling policies such as Round Robin (RR) does not perform well

• Bluetooth MAC enforces tight coupling of uplink & downlink, which leads to slot wastage

• TDD structure also restricts the packet size (1,3 or 5)

Hasan Sözer 5

Goals, Assumptions & Approaches

• Parameters of interest:

• system throughput

• packet delays

• fairness

• packet drop probability

• simplicity

• satisfying the low cost objective of Bluetooth standard.

conflicting

objectives

Hasan Sözer 6

Goals, Assumptions & Approaches (Continues...)

Criterias that an efficient scheduling policy would depend on:

• state of the queues at the Master and the Slaves

• traffic arrival process at these queues

• packet length distributions

Hasan Sözer 7

Goals, Assumptions & Approaches (Continues...)

• N queues at the Master for a piconet with N slaves

• Each slave has a queue for its connection with the Master

• Binary information is used in order to represent the state of the queues:

• 1 : has data to send 0: has no data awaiting

• State of the queue at the Slave is available at the Master (requires only 1 bit of information to transfer)

Hasan Sözer 8

Priority Policy (PP)

• There are four possibilities for the state of the queues regarding a connection:

• 1-1: Both Master and Slave have data to send

• 1-0 or 0-1: Only one side has data awaiting

• 0-0: Neither of them has data to send• PP assigns different priorities to these:

• 1-1 > 1-0 = 0-1, 0-0 is not scheduled

• It is also argued that it could be 1-0 > 0-1(*)

* Master:1 – Slave:0 > Master:0 – Slave:1

Hasan Sözer 9

K-Fairness Policy (KFP)• Beyond optimization and system throughput:

Having a strict fairness bound• qmax: Master-Slave queue pair that has

received maximum excess service (service sacrified to it)

• qmin: Master-Slave queue pair that has sacrificed maximum service to other connections

• (Services of qmax – Services of qmin) can be at most K

• When K = 0, KFP tuns out to be pure Round Robin

• In order to prevent more sacrifices: Change 1-0 into 1-1

Hasan Sözer 10

Scheduling Data in Presence of Voice

• Extend PP (to HOL-PP) & KFP (to HOL-KFP)

• Consider slot utilization by using Head-of-the-line (HOL) packets (higher utilization -> higher priority)

Hasan Sözer 11

Bluetooth SAR Policy & Possible Improvements

•Bluetooth Segmentation and Reassembly (SAR):

• naive SAR is random: assigns data packet sizes (1, 3 or 5) probabilistically.

• Instead, data arrival rates at the Master and Slave queues can be used -> Intelligent SAR (ISAR) (?):

• Initially all queues have packet size of 1

• Packet sizes change according to the differences in arrival rates at the Master and Slave

• Binary information represent high/low data rates

Hasan Sözer 12

Results & Conclusion

• Simulation results (K=500 & P=4, for 5000 TDD slots):

• KFP > PP > RR in throughput

• KFP < PP < RR in average delay (units of slots)

• KFP gives better throughput than PP with more fairness

• ISAR > SAR by means of throughput

• Keep It Simple and Stupid!

Hasan Sözer 13

Interconnecting Bluetooth-like Personal Area Networks

Godfrey Tan

MIT Laboratory of Computer Science

Hasan Sözer 14

Outline

• Conclusion

• Challenges of Interconnecting Bluetooth-like PANS & proposed solutions for each:

• Scatternet topology formation

• Packet routing

• Channel or link scheduling

Hasan Sözer 15

Scatternet Formation• Decentralized and self-healing algorithm

• Unique address for each node that are connected in a tree structure (constructed incrementally)

• Loop-free

• No packet overhead

• No periodic routing messages

• New nodes join with search announcements (root or the new node can choose among possible points of attachement)

Hasan Sözer 16

Scatternet Formation (Continues...)

0N

0*

10N-1

1*

11*

110N-

2

10*

101*

100*

1010N-3

1011*

1010*

10110N-

4

• bk = k b’s, where

b = 0 or 1

• Each node holds the portion of the address space allocated to each child

Hasan Sözer 17

Packets Relaying & Channel Scheduling

• Relaying of packets are accomplished by means of a technique that is similar to forwarding of IP packets

• makes use of longest-prefix match• Channel scheduling problem is declared to be similar to the maximal matching problem for bi-partite graphs

• An upper-bound of ceiling(d/2)*MaxDegree(*) is given for an algorithm of which details are not given

* MaxDegree = depth of the tree, d = distance in hops

Hasan Sözer 18

Conclusion

• It is declared that the algorithms are implemented in ns-2 and give good performance but simulation results are not presented

• The key idea is to construct the scatternet as a tree

• makes other problems easy to keep track of

• If the root is the one that hadle new attachements, it would have large overhead

• Enforcement of tree structure may cause deficiencies

Hasan Sözer 19

Scatternet Structure and Inter-Piconet Communication in the Bluetooth System

Manish Kalia, Sumit Garg, Rajeev Shorey

IBM India Research Laboratory

Hasan Sözer 20

Outline

• Piconet models and possible scatternet structures

• Single Piconet Model (SPM)

• Scatternet Model

• Two-Level Hierarchy of Piconets (TLP)

• Shared Slave Piconets (SSP)• Performance Comparisons & Conclusion

Hasan Sözer 21

Single Piconet Model (SPM)

• Single piconet is used even if there exists more then seven slaves

• Model uses the “Park mode”

• Timestamps are used in order to determine the period in which a slave remained parked/unparked

• Periodically, parked Slave with the oldest timestamp is unparked and active Slave with oldest timestamp is parked

• Each Slave remains unparked for the same time period

Hasan Sözer 22

Scatternet Model

• Notion of a “Communicating Group” (CG): A group of mobile devices which have frequent data transfer in between

• When forming scatternets try to make members of a CG reside in the same piconet

• Start with a SPM, structure the scatternet by collecting traffic flow patterns

• Master can observe destination addresses (Efficient policies for discovering and updating CGs are not investigated)

Hasan Sözer 23

Two-Level Hierarchy of Piconets (TLP)

• Centralized design

• Notion of root & leaf piconets

• Masters of leaf piconets periodically become slaves of the root piconet (temporary Masters can be assigned)

Hasan Sözer 24

Shared Slave Piconets (SSP)

• Decentralized structure

• A Slave in between, periodically switchs to the hopping pattern of two different Masters.

• Better load balancing & robust

• Routing is more complex

Hasan Sözer 25

Performance Comparisons & Conclusion

• Simulation results with to piconets:

• System throughput: SSP > TLP > SPM

• Average System Delays SPM >> TLP > SSP

• Scatternet allows simultaneous communication in different piconets

• In TLP leaf piconets periodically suspend communication

• SPM can be improved by considering backlogged data at the Slave queues