Leader Election

1
Leader Election

• Bismita Srichandan

2
Leader Election

• What is Leader Election
• Why Leader election is needed
• Different design topologies
• Algorithms used
• Bully Algorithm
• Ring Algorithm
• Invitation Algorithm
• References

3
What is Leader Election?

• In distributed computing, leader election is the
  process of designating a single process as the
  organizer, coordinator, initiator or sequencer of
  some task distributed among several computers
  (nodes).

4
Why Leader Election is required

• The existence of a centralized controller greatly
  simplifies process synchronization
• However, if the central controller breaks down,
  the service availability can be limited
• The problem can be alleviated if a new controller
  (leader) can be chosen
• How leader is chosen?
• Ans Different Algorithms are used for leader
  election.

5
Leader Election criteria(Chow et al, 1997)

• 1 Extrema finding - Leader election is based on
  global priority.
• 2 Preference-based - Processes in the group can
  vote for leader based on personal preference
  (locality, reliability, estimation).

6
Leader Election Algorithms for Wireless Ad Hoc
Networks10

• Two algorithms, Secure Extrema Finding Algorithm
  (SEFA) and Secure Preference-based Leader
  Election Algorithm (SPLEA).
• SEFA assumes that all elector-nodes share a
  single common evaluation function that returns
  the same value.
• In SPLEA , the scenario becomes more complicated
  because elector-nodes can have different
  preferences for a candidate-node.

7
Leader Election Algorithms for Wireless Ad Hoc
Networks

• One relaxed assumption in synchronous distributed
  systems, Asynchronous Extrema Finding Algorithm
  (AEFA) is used.
• In this we allow the topology to change during
  the election process.
• In AEFA, nodes can start the process of election
  at different times, but after topological changes
  stop long enough for the algorithm to terminate,
  all nodes agree on a unique leader.

8
Election vs. Synchronization

• Election

• Synchronization

• In election, all participants must know who owns
  the token(i.e. who is the leader).
• Election algorithms are performed only if failure
  occurs.

• In synchronization, the nontoken holders only
  need to know that they do not hold the token.
• Synchronization algorithms are designed to work
  well in absence of failures.

9
Leader Election vs. Mutual Exclusion

• Leader Election

• Mutual Exclusion

• Process may yield to others and execute normally
  as long as leader is selected.
• Concerned with fast and successful termination of
  election process.
• Result of leader election must be known to other
  processes.

• Process competes until it succeeds.
• Must ensure that no process is starved.
• Does not care which process is running in
  critical section.

10
Design Topologies

• Complete Topology
• Logical Ring Topology
• Tree Topology
• Complete Topology
• Each process in the group can reach any other
  process in the same group in one message hop.
• Assumptions taken
• All process ids are unique and known to every
  process
• Second assumption is communication network is
  reliable and only the communicating processes may
  fail.
• Process takes a known finite amount of time to
  handle a message, in other words, the process is
  considered as crashed if it doesnt respond
  within specified time-out period.
• Note Bully algorithm is based on the
  assumptions of complete topology

11
Bully Algorithm(Author Tanenbaum)

• This is an extrema-finding algorithm.
• In this algorithm, the highest-numbered process
  becomes coordinator. Thus the biggest guy in
  town always wins, hence the name Bully
  Algorithm.
• When a process notices that the coordinator is
  not responding to requests, it initiates an
  election.
• Bully algorithm makes note of time of by which
  the process should respond.
• Election is held as follows
• P sends an ELECTION message to all processes with
  higher numbers.
• If no one responds, P wins the election and
  becomes coordinator.
• If one of the higher-ups answers, it takes over.
  Ps job over!

12
Bully Algorithm
13
Bully AlgorithmProcess 6 tells 5 to stop
Process 6 wins and tells everyone
14
Logical Ring Topology

• Easy to construct
• Unique property is that a message initiated by
  any node will return to the node.

15
Ring Algorithm

• In Ring algorithm if any process notices that the
  current coordinator has failed, it starts an
  election by sending message to the first
  neighbor on the ring.
• The election message contains the nodes
  process identifier and is forwarded on around
  the ring.
• Each process adds its own identifier to the
  message.
• When the election message reaches the originator,
  the election is complete.
• Coordinator is chosen based on the highest
  numbered process.

16
Ring Algorithm
17
Ring Algorithm contd
18
Tree Topology

• To construct a logical ring structure is easy if
  the underlying network supports broadcast
  facilities.
• A tree is used as representative topological
  structure.
• Each node is considered an autonomous entity that
  exchanges messages with adjacent nodes
• A minimum-weight spanning tree(MST) is used for
  leader election in tree topology
• Gallager, Humbelt, and Spiras algorithm is based
  on searching and combining.
• It works by merging fragments, starting from each
  node and attaching level by level till it ends up
  with the MST
• A leader can be designated as the last node that
  merges and yields to the final MST.

19
Invitation Algorithm

• Invitation algorithm is a strong algorithm and it
  is basically designed for asynchronous systems.
• It eliminates possible drawbacks of Bully
  algorithm.
• Few drawbacks of Bully Algorithm
• Bully algorithm strongly uses the facts that
  timeouts can accurately detect failed processors.
• An arbitrary timing glitch (lost messages,
  overfull buffers, temporary overloads etc) , the
  bully algorithm can elect two leaders.
• Another possibility can be the node that fails
  to respond might not have failed.

20
Invitation Algorithm contd

• Each node in a group periodically checks whether
  the leader is alive or not, by sending a message
  to the leader and waits for reply.
• If the node does not reply within a timeout
  period, the node invokes a Recovery procedure.
• The recovery procedure puts node i into a
  singleton group with node i as the leader.
• periodically, each leader i calls a check
  procedure, which sends messages to every other
  node asking whether that node is a leader.
• If many nodes responds that it is a leader, node
  i pauses for some time and then calls for a merge
  procedure.

21
Invitation Algorithm contd

• Merge procedure sends message to all the
  leaders, inviting them to join a new group with
  the inviting node as leader.
• Example Group communication system in Amoeba
  uses Invitation Algorithm to reconfigure a group
  after a node crashes8.

22
References

• 1 Distributed Operating Systems
  Algorithms, Randy Chow and Theodore Johnson,
  Addison Wesley, 1997.
• 2 http//www.cs.wustl.edu/kjg/CS333_SP97/lead
  er.html
• 3 http//www.risc.uni-linz.ac.at/software/da
  j/Invitation/
• 4 http//ieeexplore.ieee.org
• 5 H. Garcia-Molina. Elections in a
  distributed computing system. IEEE Transactions
  on Computers, C- 31(1)47-59, January 1982.
• 6 http//www.cs.indiana.edu/pub/techreports/T
  R521.pdf
• 7 Distributed Systems principles and
  paradigms by Andrew S. Tanenbaum, Maarten van
  Steen, 2002.
• 8 M.F. Kaashoek and A.S. Tanenbaum, Group
  Communication in the Amoeba Distributed Operating
  System, Proc. IEEE 11th Int'l Conf. Distributed
  Computing Systems (ICDCS), pp. 222-230, 1991.

23
References

• Ring Algorithm
• 9 http//rocw.raifoundation.org/computing/BCA
  /distributedcomputing/lecture-notes/lecture-18.pdf
  
• 10Vasudevan, S. DeCleene, B. Immerman, N.
  Kurose, J. Towsley, D. Leader election
  algorithms for wireless ad hoc networks, DARPA
  Information Survivability Conference and
  Exposition, 2003. Proceedings
• 11 Optimal Leader Election Scheme for
  Peer-to-Peer Applications
• Networking, 2007. ICN '07. Sixth International
  Conference on Volume , Issue , 22-28 April 2007
• http//ieeexplore.ieee.org/xpl/RecentCon.jsp?pun
  umber4196186

24
Thank You!

• Questions