Searchlight: Won't You Be My Neighbor?

1
Searchlight Won't You Be My Neighbor?

• Mehedi Bakht, Matt Trower, Robin Kravets
• Department of Computer Science
• University of Illinois

2
Is anybody out there?
3
Is anybody out there?

• Registration services
• Foursquare, Facebook, Google Latitude
• - centralized, slow, difficult to manage across
  apps

Provides applications with absolute locations
4
Is anybody out there?

• Direct mobile-to-mobile communication
• QualComm AllJoyn, Nokia Sensor, Nintendo
  StreetPass, Sony Vita, Wi-Fi Direct
• Local, reduced latency, up-to-date,
  user-controlled

Enables applications to focus on proximity
instead of absolute location!
5
Wont you be my neighbor?

• Detection Challenges
• Encounters are unplanned and unpredictable
• Requires constant scanning
• Nodes are energy-constrained
• Requires effective duty cycling
• Global Synchronization is difficult
• Requires asynchronous solutions

?
?
?
?
?
Goal Continuous Energy-efficient Asynchronous
Neighbor Discovery
6
Energy Efficiency Duty-cycling

• Basic Discovery Idea
• Time is slotted
• Nodes selectively remain awake for a full slot
  duration
• Nodes beacon at the beginning and end of an awake
  slot
• Discovery occurs when two active slots overlap

Awake slots
7
Duty-cycled Neighbor Discovery

• Challenges
• Dealing with unsynchronized slots
• Choosing active slots
• Dealing with asymmetric duty cycles

Active Slot Selection
Awake slots
8
Slot Selection Random

• Advantage
• Good average case performance
• Disadvantage
• No bounds on worst-case discovery latency

• Birthday protocol
• Randomly select a slot to wake up in with a given
  probability

Cumulative Discovery Latency
Long tail
Is a small delay bound really necessary? Average
discovery ? Useful contact time Worst-case ?
Missed contacts
Fraction of Discoveries
Good Avg. Case Performance
Discovery Latency
9
Slot Selection Deterministic

• Disco (Sensys 2008)
• Each node selects two primes p1i and p2i
• Both nodes wake up every p1th and p2th slot (5th
  and 7th)
• Guarantees discovery in p1i x p1j slots
• U-Connect (IPSN 2010)
• Each node selects one prime pi
• Every node wakes up every pth slot and (p-1)/2
  slots every pp slots
• Overlap is guaranteed within pi x pj slots

Both Disco and U-Connect handle symmetric and
asymmetric duty cycles
10
Slot Selection Deterministic

• Prime-based
• Advantage
• Strict worst-case bound
• Disadvantage
• Poor average-case performance

• Can we get the best of both worlds
• Good average discovery latency from random
  protocols
• Good delay bound from deterministic protocols

Cumulative Discovery Latency
Disco
U-Connect
Fraction of Discoveries
Birthday
Discovery Latency
11
Searchlight

• Approach
• Have a deterministic discovery schedule that has
  a pseudo-random component
• Consider two nodes with the same (symmetric) duty
  cycles
• Insight
• Offset between slots with fixed period remains
  fixed

3 slots
Node A
Node B
B
B
B
12
Searchlight

• Approach
• Have a deterministic discovery schedule that has
  a pseudo-random component
• Consider two nodes with the same (symmetric) duty
  cycles
• Insight
• Offset between slots with fixed period remains
  fixed
• B will fall in the first ?t/2? slots of As cycle
  or A will fall in the first ?t/2? slots of Bs
  cycle

4 slots
Node A
Node B
B
B
B
4 slots
13
Searchlight

• Approach
• Have a deterministic discovery schedule that has
  a pseudo-random component
• Consider two nodes with the same (symmetric) duty
  cycles
• Insight
• Offset between slots with fixed period remains
  fixed
• B will fall in the first ?t/2? slots of As cycle
  or A will fall in the first ?t/2? slots of Bs
  cycle

4 slots
Node A
Node B
B
B
B
4 slots
14
Systematic Probing

• Technique
• Select a fixed period t (does not need to be
  prime)
• Keep one slot fixed (anchor slot)
• Add a second probe slot
• Objective is to meet the fixed/anchor slot of the
  other node
• Only need to search in the range 1 to ?t/2?
• No need to probe all ?t/2? slots all of the time
• Move around the probe slot

t
Node A
Node B
B
B
B
15
Sequential Probing

• Two slots per period t
• Anchor slot Keep one slot fixed at slot 0
• Probe slot Move around the other slot
  sequentially
• Guaranteed overlap in tt/2 slots
• Improved bound over existing protocols
• Based on the time needed to ensure a probe-anchor
  overlap
• But Probe-probe overlap should also lead to
  discovery
• Sequential scanning can result in probes
  chasing each other

1
2
3
1
2
2
3
1
2
3
Discovery through anchor-probe overlap
16
Randomized Probing

• Break the pattern of chasing
• Move the probe slot randomly (A 1-3-2 B 3-1-2)
• Pseudo-random instead of random
• Each node randomly chooses a schedule for its
  probe slot that repeats every (tt/2) slots
• Schedules of two nodes appear random to each
  other
• Advantage
• Retains the same worst-case bound
• Improves average case performance

1
3
2
1
3
1
3
2
1
3
Discovery through probe-probe overlap
17
Evaluation

• Metrics
• Fixed Energy
• All protocols operate at the same duty cycle
• Latency
• Worst-case latency bound
• Cumulative discovery latency
• Methods
• Empirical and Simulation
• Implementation
• Testbed of G1 android and Nokia N900 phones

• Comparison Protocols
• Birthday
• Disco
• U-Connect
• Searchlight Protocols
• Sequential ( Searchlight-s)
• Random (Searchlight-r)
• Scenarios
• Symmetric and asymmetric duty cycles

18
Worst-case Latency Bound

• Metric Energy Latency Product

Protocol Duty Cycle Parameters Worst-case Latency Duty Cycle Worst-case bound for duty cycle 1/x Duty-cycle for same bound
Disco p1, p2
U-Connect p
Searchlight t
19
Worst-case Latency Bound

• Metric Energy Latency Product

Protocol Duty Cycle Parameters Worst-case Latency Duty Cycle Worst-case bound for duty cycle 1/x Duty-cycle for same bound
Disco p1, p2 p1 p2
U-Connect p p2
Searchlight t t(t/2)
20
Worst-case Latency Bound

• Metric Energy Latency Product

Protocol Duty Cycle Parameters Worst-case Latency Duty Cycle Worst-case bound for duty cycle 1/x Duty-cycle for same bound
Disco p1, p2 p1 p2 4x2
U-Connect p p2 2.25x2
Searchlight t t(t/2) 2x2
21
Worst-case Latency Bound

• Metric Energy Latency Product

Protocol Duty Cycle Parameters Worst-case Latency Duty Cycle Worst-case bound for duty cycle 1/x Duty-cycle for same bound
Disco p1, p2 p1 p2 4x2 2/x
U-Connect p p2 2.25x2 1.5/x
Searchlight t t(t/2) 2x2 1.41/x
22
Symmetric Duty Cycles
Cumulative Discovery Latency
Fraction of Discoveries
Discovery Latency in Number of Slots
5 duty cycle
23
Symmetric Duty Cycles
Cumulative Discovery Latency
Fraction of Discoveries
Discovery Latency in Number of Slots
5 duty cycle
24
Symmetric Duty Cycles
Cumulative Discovery Latency
Fraction of Discoveries
Discovery Latency in Number of Slots
5 duty cycle
25
Symmetric Duty Cycles
Cumulative Discovery Latency
Fraction of Discoveries
Discovery Latency in Number of Slots
5 duty cycle
26
Symmetric Duty Cycles

• Searchlight does not have the long tail of other
  deterministic protocols
• Searchlight-R performs almost as good as Birthday
  in the average case

820
960
Fraction of Discoveries
Discovery Latency in Number of Slots
27
Expected Latency VS. Duty Cycle

• Searchlight-R performs best for all duty cycles
• Difference with other protocols increases with
  decrease in duty cycle

16
28
Expected Latency VS. Duty Cycle

• Searchlight-R performs best for all duty cycles
• Difference with other protocols increase with
  decrease in duty cycle

18
29
Handling Duty Cycle Asymmetry

• Why?
• Different energy requirements
• Different duty cycles (different values for t)
• Problem
• Anchor slots no longer have constant distance

Node A (period5)
Node B (period3)
30
Handling Duty Cycle Asymmetry

• Solution
• Restrict choice of period to primes
• Overlap of anchor slots guaranteed through
  Chinese remainder theorem
• t needs to be prime
• Worst case latency is t1 t2

Node A (period5)
Node B (period3)
31
Asymmetric (1 and 5)
Cumulative Discovery Latency

• Searchlight-R
• Worst-case latency is worse than both Disco and
  U-Connect
• Compensates for that by having best average case
  performance

82
Fraction of Discoveries
Discovery Latency in Number of Slots
32
Can we do better?

• Observation
• When slots are not fully aligned, slots of
  neighboring nodes overlap more than once within
  bound
• One overlap is sufficient for discovery!

Probe Slot 1
Probe Slot 2
Anchor Slot
Anchor Slot
33
Striping across the rounds

• Insight
• Only need to probe alternate slots
• Reduces the number of active slots by almost ½!
• Problem
• Slot alignment

Probe Slot 1
Probe Slot 2
Probe Slot 3
Anchor Slot
Probe Slot 4
Anchor Slot
34
Handling Slot Alignment

• Let the slots overflow a bit
• Extent of overlap (?) depends on
• Beacon transmission time
• Possible clock drift

1
2
3
4
5
6
Probe Slot
Probe Slot
Anchor Slot
Anchor Slot
d
35
Does it help?
Protocol Duty Cycle Parameters Worst-case Latency Duty Cycle Worst-case bound for duty cycle 1/x Duty-cycle required for same worst-case bound
Disco p1, p2 p1 p2
U-Connect p p2
Searchlight t t(t/2)
Striped Searchlight t, d t(t/4)
d amount of overflow beyond regular slot
boundary
36
Does it help?
Protocol Duty Cycle Parameters Worst-case Latency Duty Cycle Worst-case bound for duty cycle 1/x Duty-cycle required for same worst-case bound
Disco p1, p2 p1 p2 4x2
U-Connect p p2 2.25x2
Searchlight t t(t/2) 2x2
Striped Searchlight t, d t(t/4) (1d) 2x2
37
Does it help?
Protocol Duty Cycle Parameters Worst-case Latency Duty Cycle Worst-case bound for duty cycle 1/x Duty-cycle required for same worst-case bound
Disco p1, p2 p1 p2 4x2 2/x
U-Connect p p2 2.25x2 1.5/x
Searchlight t t(t/2) 2x2 1.41/x
Striped Searchlight t, d t(t/4) (1d) 2x2 (1d)/x
38
Striping and Asymmetry

• Problem
• Anchor slots no longer have constant distance
• Striping cannot be used
• Original approach
• Restrict choice of t to primes
• Worst-case bound worse than other deterministic
  protocols

39
Maintaining Constant Offset

• New approach
• Restrict value of the bigger period to an integer
  multiple of the smaller period
• Other protocols also restrict the choice of
  values for their parameters
• Only primes are allowed by Disco and U-Connect

Node A (period6)
Node B (period3)
40
Symmetric Duty Cycles
Cumulative Discovery Latency
Worst-case bound 2000 slots
Fraction of Discoveries
Discovery Latency in Number of Slots
5 duty cycle
41
Symmetric Duty Cycles
Cumulative Discovery Latency
Worst-case bound 961 slots
Fraction of Discoveries
Discovery Latency in Number of Slots
5 duty cycle
42
Symmetric Duty Cycles
Cumulative Discovery Latency
Worst-case bound 800 slots
Fraction of Discoveries
Searchlight-S
Discovery Latency in Number of Slots
5 duty cycle
43
Symmetric Duty Cycles
Cumulative Discovery Latency

• Striped probing improves bound by almost 50

Worst-case bound 440 slots
Fraction of Discoveries
Searchlight-S
Discovery Latency in Number of Slots
5 duty cycle
44
Asymmetric Duty Cycles
Worst-case bound 2266 slots
Fraction of Discoveries
Searchlight-S
Discovery Latency in Number of Slots
1-10 duty cycle
45
Asymmetric Duty Cycles
Worst-case bound 1819 slots
Fraction of Discoveries
Searchlight-S
Discovery Latency in Number of Slots
1-10 duty cycle
46
Asymmetric Duty Cycles

• Randomized probing does not have the same
  worst-case bound

Fraction of Discoveries
Searchlight-S
Discovery Latency in Number of Slots
1-10 duty cycle
47
Restricted Randomized Probing

• Randomization across tA/2 could delay discovery
• Restrict randomization based on smallest t
• Impact
• Same bound as sequential for asymmetric case
• No effect on symmetric case

Node A (period16)
1
2
3
Node B (period8)
48
Implementation Issues

• Larger slot size
• Off to on transition of the Wi-Fi card takes
  seconds
• Faced same problem with three different phones
• Android G1/G2, Nokia N900, Nexus-S
• Duration of a slot was 3-4 seconds
• U-Connect Disco was implemented on sensors with
  slot size in the order of hundreds of
  milliseconds

49
What should I use?

• Mostly symmetric duty cycles
• Searchlight with restricted randomized striped
  probing
• For any two nodes with the same duty cycle
• Best average and best worst-case bound
• For any two nodes with different duty cycles
• Almost best average and best worst-case bound
• Very diverse duty cycles
• Searchlight with symmetric striped probing
• Has slightly better average discovery latency

50
Searchlight Won't You Be My Neighbor?

• http//mobius.cs.uiuc.edu