AdHoc Network Scheduling

1
Ad-Hoc Network Scheduling

• Bluetooth Scatternets Scheduling Scheme
• Based on
• Bluetooth scatternets an enhanced adaptive
  scheduling scheme
• by Simon Baatz, Matthias Frank, Carmen Kuhl,
  Peter Martini, Christoph Scholz

2
Agenda

• Introduction
• Bluetooth Overview
• Piconet Scheduling
• Challenges In Scatternet Support
• Credit Based Scatternet Scheduling
• Simulation Results
• Summary

3
Introduction

• Bluetooth technology supplies a convenient
  solution for mobile connectivity.
• The basic Bluetooth network (piconet) supports
  only a limited number of devices.
• Scatternets are multi-hop networks comprising
  multiple piconets.
• A scheduling protocol is needed in piconets
  scatternets.

4
Agenda

• Introduction
• Bluetooth Overview
• Piconet Scheduling
• Challenges In Scatternet Support
• Credit Based Scatternet Scheduling
• Simulation Results
• Summary

5
Bluetooth Overview - Piconets

• The Bluetooth piconet consists of 1 master device
  and up to 7 slave devices.
• Slaves cannot communicate directly with each
  other.
• The master defines the timing in the piconet.

6
Bluetooth Overview - Scatternets

• A scatternet is formed by interconnecting
  multiple piconets.
• Piconet interconnection is reached by devices
  (connecting devices) participating in multiple
  piconets.

7
Bluetooth Overview SCO ACL links

• Bluetooth specifies synchronous connection
  oriented (SCO) links and asynchronous
  connectionless links (ACL).
• SCO communication is organized in previously
  reserved slots.
• Access to ACL with the remaining bandwidth is
  controlled by a time division duplex (TDD)
  scheme.
• We will focus on ACL only.

8
Agenda

• Introduction
• Bluetooth Overview
• Piconet Scheduling
• Challenges In Scatternet Support
• Credit Based Scatternet Scheduling
• Simulation Results
• Summary

9
Piconet Scheduling

• The master may begin to send a packet in even
  numbered slots only.
• The addressed slave may send in the odd numbered
  slot after the masters packet.
• Packets must occupy an odd number of slots (1,3
  or 5 slots).
• A master can send a POLL or NULL packet to a
  slave in order to give it a chance to send a
  packet.
• A slave can answer with a NULL packet.

10
Piconet Scheduling
11
Low Power Modes

• Why should we use low power modes?
• Energy is a very important resource in ad-hoc
  networks.
• Moving devices to low power modes can free places
  for other devices in the piconet.

12
Low Power Modes

• Park mode
• The slave cannot participate in the piconet.
• Parking/Unparking a slave has an overhead on
  system performance.
• Hold mode
• The slave is inactive for a fixed time interval.
• Sniff mode
• The slave periodically wakes up to listen to
  master tranmission.

13
Low Power Mode (Sniff Mode)

• Sniff mode is used in order to reduce the duty
  cycle on a link between 2 devices.
• Every 2 devices negotiate a specific slot (sniff
  slot) where communication may start.
• If no communication starts in the sniff slot,
  both devices may go to low power mode.
• If communication starts, it ends after one device
  ends it.

14
Low Power Mode (Sniff Mode)
15
Agenda

• Introduction
• Bluetooth Overview
• Piconet Scheduling
• Challenges In Scatternet Support
• Credit Based Scatternet Scheduling
• Simulation Results
• Summary

16
What is the main problem?

• The connecting devices need to share time between
  piconets.

17
Challenges In Scatternet SupportSwitching
Piconets

• Different piconets may use different timing.
• Devices will typically use a single Bluetooth
  transceiver.
• Therefore, a device has to switch between
  piconets.
• When switching, some slots may be unavailable for
  communication.

18
Challenges In Scatternet SupportScheduling
Switches

• A device may enter a low-power mode in its
  current piconet and participate in different
  piconet.
• Devices must coordinate about when to switch
  piconets.
• A device can use only one link at a time.

19
Challenges In Scatternet SupportAdaptivity

• In the previous example only 1/3 of the airtime
  can be used for a link (even if other links are
  idle).
• Changing the schedule of a single link may affect
  the whole scatternet.
• Frequent recalculations are needed in order to
  achieve adaptivity.

20
Challenges In Scatternet SupportClock Drift

• The timing in a piconet is determined by a clock
  in the master device.
• According to Bluetooth standards, the allowed
  clock drift may cause 2 piconets to drift against
  each other up to 2 slots every 30 seconds.

21
Agenda

• Introduction
• Bluetooth Overview
• Piconet Scheduling
• Challenges In Scatternet Support
• Credit Based Scatternet Scheduling
• Simulation Results
• Summary

22
So, what is the solution?

• Calculating piconet schedules for the whole
  scatternet requires too many resources.
• Using only local communication schedule may waste
  much bandwidth.
• Solution using a credit based scatternet
  scheduling.

23
Credit Based Scatternet SchedulingPresence
Points

• Presence points are defined for each
  inter-piconet link.
• In a presence point, communication between master
  and slave over the corresponding link may start.
• Presence points may be mapped directly onto sniff
  mode (almost no changes are required).

24
Credit Based Scatternet SchedulingCredit Scheme

• A device has to decide when to abort an ongoing
  sniff event in order to use an upcoming sniff
  event.
• The decision will be taken according to a
  priority scheme based on leaky bucket traffic
  shaping.

25
Credit Based Scatternet SchedulingCredit Scheme
(cont)

• Each device manages a priority for each of its
  links.
• Two devices on both ends of a link may assign
  different priorities for the same link.
• Each link of a device has a credit account.
• Each credit represents the use of 1 slot of the
  corresponding link.

26
Credit Based Scatternet SchedulingCredit Scheme
(cont)
27
Credit Based Scatternet SchedulingCredit Scheme
(cont)

• If a slot is not used by any link, a temporary
  account is debited.
• Since one credit is leaving the system, one
  credit is entering the system on every slot by
  increasing the temporary account.
• When the temporary account holds n credits (n is
  the number of links), these credits are divided
  between the links accounts.

28
Credit Based Scatternet SchedulingCredit Scheme
(cont)

• acx account of link x
• actemp temporary account

29
Credit Based Scatternet SchedulingCredit Scheme
(cont)

• An ongoing sniff event is aborted in order to use
  another links upcoming sniff slot if the latter
  has a higher number of credits.
• This decision is not agreed on between the 2
  devices participating in the sniff event.

30
Credit Based Scatternet SchedulingCredit Scheme
(cont)

• A large amount of credits states that the link
  wasnt active lately.
• A link has more credits relative to another link
  if it has been treated unfairly in the past when
  compared to the other.

31
Credit Based Scatternet SchedulingCredit Scheme
(cont)

• Other cases in which a sniff event is aborted
• If another link is about to reach its poll
  interval. It ensures an upper bound on the time
  between 2 polls on the same link.
• If the slave sent a NULL after receiving POLL
  from the master.
• If timeout has been reached.

32
Credit Based Scatternet SchedulingSwitch
Threshold

• Problem The basic credit scheme may lead to
  frequent switches between links.
• Solution
• It is desirable to minimize the number of switch
  events.
• A link will be served at least for a minimum
  number of slots.

33
Credit Based Scatternet SchedulingSwitch
Threshold - Starvation

• Guaranteed service time may lead to starvation.

34
Credit Based Scatternet SchedulingSwitch
Threshold - Solution

• bool abort_sniff_event(from,to)
• if (acto gt acfrom Nswitch_th)
• return true
• else
• return false
• acto credits of the link which has a sniff slot
  during the sniff event.
• Acfrom credits of the link which is currently
  in a sniff event.
• Nswitch_th - constant

35
Credit Based Scatternet SchedulingRedistribution
of Credits

• Problem
• A device may have links which dont use their
  bandwidth.
• These links will collect credits but never use
  them.
• The credits will be used only for frequent
  POLL-NULL sequences.

36
Credit Based Scatternet SchedulingRedistribution
of Credits

• Solution
• A link that does not fully use its bandwidth
  share must release the unused bandwidth to other
  links.
• This approach deals only with the links inside a
  Bluetooth device and does not consider end to end
  streams.

37
Credit Based Scatternet SchedulingRedistribution
of Credits cont

• There are 2 approaches regarding the
  redistribution of credits
• Absolute All positive credits will be divided
  between other links.
• Relative Only the difference of credits between
  2 links is meaningful.

38
Credit Based Scatternet SchedulingRelative
Redistribution of Credits

• The amount of credits to be distributed is
• acred credits of the link to be redistributed
• acmin credits of a link with minimal amount of
  credits
• n number of links of the device

39
Credit Based Scatternet SchedulingRedistribution
of Credits - Example
40
Credit Based Scatternet SchedulingAdaptive
Presence Point Density (APPD)

• Problem
• A link that has very low load is activated too
  frequently.
• Links may waste their credits for unsuccessful
  sniff slots or POLL-NULL sequences.
• Solution
• The density of sniff slots is reduced
  exponentially over time.

41
Credit Based Scatternet SchedulingAdaptive
Presence Point Density (APPD)

• Each device manages an internal Tsniff for each
  link.
• Tsniff is initialized to Tsniff.
• If a transmission occurs on the sniff slot,
  Tsniff Tsniff.
• If no transmission occurs on the sniff slot,
  Tsniff Tsniff 2.
• Tsniff cannot exceed Tpoll.

42
Agenda

• Introduction
• Bluetooth Overview
• Piconet Scheduling
• Challenges In Scatternet Support
• Credit Based Scatternet Scheduling
• Simulation Results
• Summary

43
Simulation ResultsCredit Scheme

• Scenario
• 3 piconets with 5 streams.
• All links of node F are saturated.
• Piconets are synchronized.
• Each link of node F has a different Tsniff value.

44
Simulation ResultsCredit Scheme

• The total throughput of the node is 147.2 kbs.
• Each of the 5 links gets 29.44 kbs (1/5 of total
  throughput).

45
Simulation ResultsCredit Scheme

• Links with short sniff periods have shorter sniff
  events.

46
Simulation ResultsCredit Redistribution

• All links of node F have the same Tsniff.
• Nswitch_th 50
• This leads to less frequent piconet switches.
• Since the service cycle is longer, other links
  can collect more credits.
• All streams saturate their links and then reduced
  at 5,10,15,20,25 seconds respectively.

47
Simulation ResultsNo Credit Redistribution

• The freed bandwidth isnt reallocated.
• Most of the airtime allocated to idle links is
  consumed by extensive polling.

48
Simulation ResultsCredit Redistribution
49
Simulation Results Adaptivity

• Scenario
• Packet transmission is assumed to be error-free.
• Retransmissions are performed only in case of
  unsuccessful sniff slots.
• TCP packets have a size of 1500 bytes.

50
Simulation ResultsAdaptivity

• A FTP stream is sent from device A to device E
  via device C.
• A Video stream at 256 kbps is sent from device A
  to device D via device B.

51
Simulation Results Adaptivity

• The average goodput doesnt suffer when
  increasing Tsniff.
• When device C is not served by device A, it sends
  data to device E.

52
Simulation Results Adaptivity

• A FTP stream is sent from device A to device E
  via device C.
• A Video stream at 64 kbps is sent from device A
  to device D via device B.
• Since the video stream has a low bandwidth, link
  0 will be partially used.

53
Agenda

• Introduction
• Bluetooth Overview
• Piconet Scheduling
• Challenges In Scatternet Support
• Credit Based Scatternet Scheduling
• Simulation Results
• Summary

54
Summary

• Credit based scheduling has the following
  advantages
• It enables a fair time sharing between all links
  of a device.
• It reduces piconet switches.
• Adaptivity there is almost no unused bandwidth
  caused by links with low load.

55
References

• Bluetooth scatternets An Enhanced Adaptive
  Scheduling Scheme http//ieeexplore.ieee.org/Xpl
  ore/Toclogin.jsp?url/iel5/7943/21922/01019324.pdf
  ?isNumber21922prodCNFarnumber1019324arSt782
  ared790arAuthorBaatz2CS.3BFrank2CM.3BK
  uhl2CC.3BMartini2CP.3BScholz2CC.
• Adaptive Scatternet Support for Bluetooth using
  Sniff Mode - http//web.informatik.uni-bonn.de/IV/
  Mitarbeiter/baatz/LCN2001_rp.pdf
• Scatternet Structure and Inter-Piconet
  Communication in the Bluetooth System
  http//www.cs.bilkent.edu.tr/korpe/resource/paper
  s/kalia00scatternet-bluetooth.pdf
• Specification of the Bluetooth System (Core) -
  http//www.bluetooth.com/pdf/Bluetooth_11_Specific
  ations_Book.pdf

56
Questions
57
Thank You !