Student research opportunities

Fault-tolerant Data Services for High Performance Computing on Clouds

Project Code: CECS_631

This project is available at the following levels:
Honours, Masters, PhD

Keywords:

Service-oriented architecture, high performance computing, cloud computing, data fabrics

Supervisor:

Dr Peter Strazdins

Outline:

The Service-oriented Architecture approach has been successfully applied to large-scale grid applications (which are `embarrassingly parallel'), such as in computational finance. The `embarrassingly parallel' nature of these applications is exploited, yielding a friendly parallel programming paradigm which is naturally fault-tolerant and efficient under (dynamically) heterogeneous processing conditions. The HPNumSOA project at ANU has extended this approach to general parallel applications through the use of a `data service' (see the paper in the links below). However, the desired goal of (efficiently supported) fault-tolerance in this data service has not yet been addressed in this work.

Goals of this project

This project will investigate approaches to introduce fault-tolerance in such an extended SOA framework, under the constraints of memory scalability and minimal impact on performance. Particular issues of when such a framework is implemented on a public cloud, such as data transfer time and variability in latency, will also be investigated.

Requirements/Prerequisites

An Honours degree in Computer Science or equivalent.

Student Gain

This project is an extension on an existing project with Platform (TM) Computing, with opportunities to continue the collaboration. An Internship with Platform Computing (Beijing or Toronto) may be offered. There will also be access to a moderate-sized scientific cloud infrastructure through the adjoining NCI National Facility.