GeoQuorums

1
GeoQuorums

• Implementing Atomic Memory in
• Mobile Ad Hoc Networks
• Seth Gilbert
• Shlomi Dolev, Nancy Lynch,
• Alex Shvartsman, Jennifer Welch
• DISC October 3rd 2003

2
Distributed Shared Memory
Read
Write(7)
Write(0)
3
Mobile Ad Hoc Networks

• No fixed infrastructure
• e.g. ZebraNet
• GeoSensor
• GPS
• Synchronized Clock

4
GeoCast

• send(message, location)
• See
• Navas and Imielinski (97)
• Camp and Liu (2002)

5
Quorum Based Algorithms

• See (for example)
• Gifford (79) and Thomas (79)
• Upfal and Wigderson (85)
• Attiya, Bar-Noy, and Dolev (90/95)

Write(7)
6
Problems

• Static
• Potentially unbounded and dynamic universe of
  nodes?
• Mobility
• How do you find the nodes?
• Location
• What if the members of a quorum become dispersed?

7
Reconfiguration??

• See (for example)
• RAMBO (DISC 2002) RAMBO II
• (Lynch and Shvartsman Gilbert, Lynch, and
  Shvartsman)

8
Problems

• Static
• Potentially unbounded and dynamic universe of
  nodes?
• Mobility
• How do you find the nodes?
• Location
• What if the nodes are far apart?
• Reconfiguration

9
Focal Points

• busy traffic intersection
• water source
• cafeteria
• lounge

10
Problems

• Static
• Potentially unbounded and dynamic universe of
  nodes
• Mobility
• How do you find the nodes?
• Location
• What if the nodes are far apart?
• Reconfiguration

11
New Problem

• At most f focal points become empty
• How do we respond to failed focal points?

12
Replication
read
13
GeoQuorums Approach

• Low Level Focal Point Emulator
• Implement focal points
• Simulate data objects at fixed locations
• High Level Operation Manager
• Use fixed data objects to implement atomic memory

14
Focal Point Emulator

• Operations
• put
• updates tag/value
• get
• retrieves tag/value
• confirm
• confirms a write is complete
• recon-done
• reconfiguration is complete

15
Focal Point Emulator

• Participating mobile nodes are close together
• Assume
• Local Totally Ordered Broadcast
• Replicated State Machine
• Tolerating joins/leaves

16
Operation Manager

• Write
• Choose unique tag
• use clock, pid
• put tag/value to quorum
• Read
• get latest tag/value from quorum
• (put tag/value to quorum)

mobile host
focal point data object
17
Reconfiguration

• Switch quorums of focal points
• Limited to predetermined configurations
• Primarily for performance
• Also for fault tolerance
• Simple
• Two phase operation
• get value from old configuration
• put value in new configuration

18
Geographic Quorums
focal point
cluster
mobile host
19
Safety and Performance

• At most f focal points fail
• Arbitrary message delays
• We prove that GeoQuorums implements linearizable
  read/write objects
• Good latency in partially synchronous executions

20
Conclusions

• Introduce a new algorithm for atomic memory in
  mobile, ad hoc networks
• Develop notion of focal points
• Bridge gap between static and mobile
• Use static algorithms in mobile ad hoc setting

21
The End

• (Questions?)

22
Backup Slides
23
GeoCast Service
geo-send
geo-send
geo-rcv
geo-rcv
Operation Manager
Focal Point Emulator
Mobile Host
GeoSensor
read-ack write-ack recon-ack
read/write recon
lbcast-send
lbcast-rcv
LBCast Service
Client
24
Open Problems

• Focal Points
• How to choose focal points?
• Self-stabilizing focal points?
• Reconfiguration
• When is it good to reconfigure?
• How to choose configurations?

25
Open Problems

• Focal Point ? Virtual Node
• Generalize not just for read/write objects
• Other distributed algorithms?
• Hybrid Networks