Center for Advanced Vehicular Systems - Cooperative Computing - Computational Vortals for Next Generation Scalable Computing

Home -> Research Groups -> Cooperative Computing -> Projects -> Computational Vortals for Next Generation Scalable Computing

Project

Computational Vortals for Next Generation Scalable Computing
Grant Number EIA0103594

In Collaboration with:
National Science Foundation

PI: Tomasz Haupt
Team Members: Sheikh Ghafoor

Description:

Clusters are becoming ubiquitous, cost effective means for large-scale computation at most universities and industrial settings. Clusters of desktop computers connected through high-speed network have become an alternative to vector and massively parallel supercomputers (MPP). Scheduling of processes onto processors of a parallel machine has always been an important and challenging area of research. The research is challenging because of the numerous factors involved in implementing a scheduler. Some of these influencing factors are: the parallel workload, the presence of any sequential and/or interactive jobs, the characteristics of the native operating system, hardware, network interface, etc. The recent shift towards the adoption of clusters for cost effective parallel computing makes the design of an efficient scheduler even more crucial and challenging. Traditional solutions that have been used in conventional parallel systems are not adequately tuned to handle the diverse workloads and performance criteria required by cluster environments.

One of the major assumptions of current parallel job schedulers about workloads is that the jobs are rigid, meaning that an application doesn't change resource requirements (in terms of number of processors) during execution. In spite of lot of research systems with rigid jobs competing for resources are often under-utilized. Adaptive parallel applications(applications that can change number of processor during execution) promises better resource utilization as resource can be allocated to applications as and when needed, and also applications can be expanded to utilize idle resources whenever available. An adaptive application is defined as one that changes
resource requirements during execution. The change in resource requirements may be triggered by
the application itself, due to the nature of the employed algorithms, or it might be triggered
by events external to the application, such as changes in hardware availability, applications of
higher priority requiring more resources, and others. Current job schedulers and resource
management systems are unable to handle adaptive applications efficiently.

The objective of the proposed research is to develop a system for next generation parallel
adaptive applications. The centerpiece of the system is a scheduler for cluster of workstations,
which approximates a "grid environment".

Vortal Design

The Vortal project has three major component.

Resource Negotiation Protocol for communication between adaptive applications and scheduler.
A Resource Management System which will include an scheduling subsystem capable of scheduling adaptbve parallel application to optimize an objective function.
A scheduling subsystem

Learn more

Sheikh Ghafoor, Tomasz Haupt, and Satya Gosula, "A Resource Negotiatopn Protocol for Adaptive Parallel Application", submitted to High Performace Computing Symposium (HPC 2005), San Diego, April 2005 (accepted)
Satya Gosula and Sheikh Ghafoor, "A Resource Negotiatopn Protocol for Adaptive Parallel Application: Version 1.0, User Guide"
Scheduling Adaptive Parallel Application on Cluster

Status of the Project

Resource Negotiation Protocol has been developed and tested.
Prototype Resource Management System is under development and expected tobe complete by December 2004.
Work on scheduling subsystem in ongoing