Comparison between Suzuki Kasamis and Raymonds Tree Algorithm

1
Comparison between Suzuki Kasamis and Raymonds
Tree Algorithm

• -Sagar Panchariya

2
Introduction

• Token Based Mutual Exclusion Algorithms
• A unique token is shared among all sites. A
  site is allowed to enter critical section if it
  possesses the token.
• Types of Token based algorithm
• Broad Cast
• Non BroadCast

3
Suzuki Kasamis Algorithm

• If a site needs to enter into critical section
  and if does not have a token it broad cast
  request to all the other site.
• On receiving request if a site has token if the
  token is idle, i.e. site is not executing
  critical section it sends the PRIVILEDGE message
  containing the token the requesting site.
• Each site contains a requesting site array
  Rrequesting sequences. Token maintains LN
  arrayexecuted sequences and token queue

4
Raymonds Tree Algorithm

• Site does not broad cast it sends a request along
  the directed edge to its holder.
• Site can send request along the directed edge
  only once. Every site maintains a request array
  of Processes requesting CS.
• Token is released when the token is idle with the
  process and send with a PRIVILEDGE message along
  the edge.
• If the sites id is top of its request Q then the
  site can enter critical section.

5
Performance Parameters for Mutual Exclusion
Algorithm

• Number of messages per CS entry
• Synchronization delay Time required when one
  site leaves CS and another enter CS.
• Response Time Is the time interval a site waits
  its CS execution to be over after request has
  been sent.
• Throughput1/(sdE), where sd is average
  synchronization delay, E is average CS execution
  time

6
Parameters I kept as fixed while measurements

• Number of CS executions100
• Message delay1-9
• CS execution time1-9
• Parameters I varied
• Number of nodes 30,50,100
• load (number of process simultaneously requesting
  CS). 1p,1/2p,full load.

7
(No Transcript)
8
Raymonds Staright-Line
For a straight line when all the nodes are
requesting simultaneously, for execution of all
the critical section the number of messages
required is 2N However if few number of nodes are
requesting at different states the number of
messages vary depending on the location of
requesting nodes in the chain. For half load
conditions it is like reducing load due to the
random selector used for selecting requesting
process.
9
Raymonds Star
For Full load condition 101 nodes require approx
400 messages for half load and other nodes
testing like 51,31 this varies because of the
setup and the random nature of selector. This
goes for all the other tree topologies.
10
Raymonds Tree 2 depth (30-70)
11
Raymonds Tree 2depth(70-30)
12
Conclusion

• In Raymond's Tree algorithm under full load the
  number of messages exchanged per critical section
  is approximately 4.
• In Suzuki Kasamis Algorithm the number of
  messages requesting critical section linearly
  increases the number of messages.
• Synchronization delay and throughput are
  dependent on the average executes critical
  section and average message delay

13
Future work

• More research has to be done on Raymond's Tree
  topologies and their behavior around half load
  condition to find out the ideal topology.

14
Thank you