A cloud scheduler handling performance variability of machine instances

From Master Projects
Jump to: navigation, search


A cloud scheduler handling performance variability of machine instances
status: finished
Student name: student name::Rocco Santese
Dates
Start start date:=2014/02/01
End end date:=2014/07/31
Supervision
Supervisor: Thilo Kielmann
Second supervisor: Ana-Maria Oprescu
Second reader: has second reader::Ana-Maria Oprescu
Company: has company::Vrije Universiteit Amsterdam
Thesis: has thesis::Media:Thesis.pdf
Poster: has poster::Media:Posternaam.pdf

Signature supervisor



..................................

Abstract

Execution of bags of tasks on homogeneous cloud environments may not be optimal performancewise,as different machine instances of the same type may have different performances, due to factors which are external to the bag of tasks itself. Such external factors are mainly attributed to concurrent usage of the same machine by other users and to malfunctioning or underperforming hardware. Unfortunately, performance variability leads to a lessthan optimal execution of the bag of tasks both in terms of time and budget. Since cloud users are unable to act upon a specific machine instance modifying or correcting its behavior, a clever policy for identifying and releasing underperforming machines as well as scheduling tasks is required for an optimal execution. This thesis proposes a Bag of tasks scheduler that checks the performances of all of the acquired machines during runtime. For this purpose the scheduler will use a sample of tasks of similar complexity and execution time, to identify which machine instances are underperforming compared to the others. The release of underperforming machines instances and the acquisition of new machines instances, are handled by the scheduler in order to maximize execution time and reduce budget expenses.