Sequential Execution paradigm for a Cloud

1
Sequential Execution paradigm for a Cloud

• Velidi Padmini
• 200707031

2
Introduction

• Cloud does not answer all questions
• All problems are not MapReducable
• How can a Sequential Computation be done on a
  Cloud?
• How can Sequential programs be parallelised?
• It is not just Map-Reducethen, Whats it?

3
Revised Cloud Architecture

• 
  
  

4

• 
  

5

• The master node determines number of nodes and
  clusters as per Amdahls law.
• Amdahls law
• determines the speedup of using parallel
  processors on a problem, versus using only one
  serial processor.
• where p is part of program that can be
  parallelised
• Speedup (s p ) / (s p / N )
• Where s is amount of time spent on serial program
  and
• p is amount of time spent on parallel
  parts of program
• N is number of processors
• How do we do that parallelization????

6
Design Approaches

• Parallelize existing sequential algorithm or
  modify parts of the existing serial algorithm
  which can be fully parallelized.
• Design completely new parallel algorithm
• Design new parallel algorithm from the existing
  parallel algorithm

7
Design Issues

• For any given serial problem, it can be
  parallelized in 4 steps
• Partitioning
• Communication
• Agglomeration
• Mapping

8

• Data decomposition

• Functional decomposition

9

10

• A dependence exists between program statements
  when the order of statement execution affects the
  results of the program.
• A data dependence results from multiple use of
  the same location(s) in storage by different
  tasks.
• Dependencies are important to parallel
  programming because they are one of the primary
  inhibitors to parallelism.

11


Example Array Processing

• serial

• parallel

12
Architectural Issues

• Main master allocates tasks to the sub-masters
  present in each cluster.
• Sub-master takes care of map and reduce tasks
  within a cluster.
• Along with output file, intermediate log file is
  also transferred from cluster to cluster.
• A copy of both the files reside with the master
  as well.
• Master stores its log files at sub-masters as
  well so as to deal with master failures.

13

• Thank You

Thank You