Election Algorithms

Topics
r Issues
r Detecting Failures
r Bully algorithm
r Ring algorithm
 Readings
r Van Steen and Tanenbaum: 5.4
r Coulouris: 11.3
 Election Algorithms
r Remember using Lamport clocks for total order
r Can you think of another way to do this?
r It turns out that you can use a sequencer.
m All operations go to a sequencer
m The sequencer assigns numbers to each message before
the message goes to each replica
m What if the sequencer goes down?
 Election Algorithms
r Many distributed algorithms require a
process to act as a coordinator.
r The coordinator can be any process that
organizes actions of other processes.
r A coordinator may fail
r How is a new coordinator chosen or
elected?
 Election Algorithms
 Assumptions
 Each process has a unique number to distinguish
them.
 One process per machine (which suggests that
an IP address can be the unique identifier)
 Processes know each other’s process number
 Processes do not know which ones are currently
up and which ones are down.
 General Approach
 Locate the process with the process with the
highest process number and designate it as the
coordinator.
 Election algorithms differ in how they do this.
 Issues in Dealing with
Coordinator Failure
m Detecting Failure
• Any node might detect failure first
• Multiple processes might detect failure at once.

m Election
• Must run without coordination
• Must deal with arbitrary process failures
• All nodes must agree on when election is over and who
the new coordinator is.
 Detecting Failures
r Timeouts are used to detect failures
T = 2Ttrans + Tprocess
• Where Ttran is maximum transmission delay and Tprocess
represents the maximum delay for processing a
message.
r If a process fails to respond to a message
request within T seconds then an election
is initiated.
 Bully Algorithm
 When a process, P, notices that the
coordinator is no longer responding to
requests, it initiates an election.
 P sends an ELECTION message to all processes
with higher numbers.
 If no one responds, P wins the election and
becomes a coordinator.
 If one of the higher-ups answers, it takes over.
P’s job is done.
 Bully Algorithm
r When a process gets an ELECTION
message from one of its lower-numbered
colleagues:
m Receiver sends an OK message back to the
sender to indicate that he is alive and will take
over.
m Receiver holds an election, unless it is already
holding one.
m Eventually, all processes give up but one, and
that one is the new coordinator.
m The new coordinator announces its victory by
sending all processes a message telling them
that starting immediately it is the new
coordinator.
 Bully Algorithm
 If a process that was previously down
comes back:
 It holds an election.
 If it happens to be the highest process
currently running, it will win the election and
take over the coordinator’s job.
 “Biggest guy” always wins and hence the
name “bully” algorithm.
The Bully Algorithm (Example)

r The bully election algorithm

r Process 4 holds an election
r Process 5 and 6 respond, telling 4 to stop
r Now 5 and 6 each hold an election
The Bully Algorithm (Example)
d) Process 6 tells 5 to stop
e) Process 6 wins and tells everyone
 Bully Algorithm
Analysis
r Best case
m The node with second highest identifier
detects failure
m Total messages = N-2
• One message for each of the other processes
indicating the process with the second highest
identifier is the new coordinator.
r Worst case
m The node with lowest identifier detects failure.
This causes N-1 processes to initiate the
election algorithm each sending messages to
processes with higher identifiers.
m Total messages = O(N2)
 Bully Algorithm Discussion
r How many processes are used to detect a
coordinator failure?
m As many as you want. You could have all other
processes check out the coordinator.
r It is impossible for two processes to be
elected at the same time.
 Ring Algorithm
r Use a ring (processes are physically or logically
ordered, so that each process knows who its
successor is).
r Algorithm
m When a process notices that coordinator is not functioning:
• Builds an ELECTION message (containing its own process
number)
• Sends the message to its successor (if successor is down,
sender skips over it and goes to the next member along the ring,
or the one after that, until a running process is located).
• At each step, sender adds its own process number to the list in
the message.
 Ring Algorithm
r Algorithm (continued)
m When the message gets back to the process that
started it all:
• Process recognizes the message that contains its own
process number
• Changes message type to COORDINATOR
• Circulates message once again to inform everyone else:
Who the new coordinator is (list member with highest
number); Who the members of the new ring are.
• When message has circulated once, it is removed.
• Even if two ELECTIONS started at once, everyone
will pick same leader since node with highest
identifier is picked.
 Ring Algorithm

Initiation:
1. Process 4 sends an
ELECTION message to
its successor (or next
alive process) with its ID
 Ring Algorithm

Initiation:
2. Each process adds its own
ID and forwards the
ELECTION message
 Ring Algorithm contd…

Leader Election:
3. Message comes back to initiator, here
the initiator is 4.
4. Initiator announces the winner by
sending another message around the ring
 Ring Algorithm Analysis

• At best 2(N-1 ) messages are passed

• One round for the ELECTION message
• One round for the COORDINATOR
• Assumes that only a single process
starts an election.
• Multiple elections cause an increase in
messages but no real harm done.
 Summary
r Synchronization between processes often
requires that one process acts as a
coordinator.
r The coordinator is not fixed.
r Election algorithms determine the
coordinator.