120 likes | 293 Views
Lamport’s Scalar Clocks. Shiva Bottu. Organization. Introduction High level Implementation Details Experimentation Results Further Research. Introduction. Lamport’s logical clocks are introduced to capture causality relation in a distributed system.
E N D
Lamport’s Scalar Clocks Shiva Bottu
Organization Introduction High level Implementation Details Experimentation Results Further Research
Introduction Lamport’s logical clocks are introduced to capture causality relation in a distributed system. Each process Pi has a logical clock Ci and for a particular event a in P, it assigns an integer value Ci, which is timestamp of the event. Timestamps are monotonically increasing.
High level Implementation Details • There are 2 implementation rules that should be followed • Before executing event, update timestamp value Ci:=Ci+d • When sending a message, attach the timestamp. When message is received by a process, timestamp of receive event is calculated as maximum of the two timestamps (timestamp of the process and timestamp received in the message) Ci:=max(Ci,Cmsg)
Experimentation • Number of processes are varied. • Random flood algorithm is used to exchange messages between processes • Part 1: • Total number of updates are plotted against number of processes • Average number of updates per process against number of processes • Part 2: • Number of updates and Number of messages against number of processes • Number of processes against average number of messages required for each process
Part 1(a):Total clock updates against Number of processes Number of clock updates Number of Processes
Part 1(b):Average clock updates per process against Number of processes Average clock updates Number of Processes
Part 2(a): No. of Processes VS clock updates & No. of Processes VS No. of messages required clock updates & Messages Number of Processes
Part 2(b): No. of Processes VS Average no. of messages per process No. of Messages per process Number of Processes
Results From part-1(a), it is observed that the number of updates increase with increase in number of processes From part-1(b), average number of updates per process increase with increase in number of processes From part-2(a), we can observe that number of clock updates and number of messages increase with increase in no. of processes From part-2(b), we can observe that average number of messages required per process increases with increase in number of processes
Further Research To change the base algorithm from random flood to other topologies and test with the same process. Running with high load of processes.
References http://en.wikipedia.org/wiki/Lamport_timestamps http://cnlab.kaist.ac.kr/~ikjun/data/Course_work/CS642-Distributed_Systems/papers/lamport1978.pdf