Economic and On Demand Brain Activity Analysis on

Global Grids

A case study

Need for grid

Two major problems commonly observed in scientific disciplines:

scientific data The distribution of knowledge and technologies

Cont..

One such scientific discipline: Brain science

The analysis of brain activity data gathered from the MEG (Magnetoencephalography) instrument is an important research topic in medical science

Introduction

computational grids :• Aggregations of such distributed resources, called

computational grids.

Biological science:• Brain Activity is one such application

• Brain activity is measured by the Magnetoencephalography (MEG)

• measures the magnetic fields generated by the electrical activity in the brain.

Brain Activity Analysis - Advantage of MEG

MEG instrument: consists of a number of sensors which record

information about brain activity MEG helmets with over 200 sensors detect magnetic brain fields by means of a sensitive

transducer technology called Superconducting Quantum Interference Device (SQUID)

limitation The high cost of equipment There are only limited numbers of MEG

instruments around the world For example, a 64-sensor MEG instrument

would produce 0.9GB of data over a period of an hour.

Such a task generates 7,257,600 analysis jobs and would take 102 days on a commodity computer with a PentiumIII/500MHz processor and 256MB of memory.

Grid-based Analysis Model

NeuroGrid project aims to• convert the existing brain activity analysis

application into a parameter sweep application for executing jobs

• which perform wavelet cross-correlation analysis for each pair of sensors in parallel on distributed resources

A Model for Brain Activity Analysis on Global Grids.

steps:

1. medical staff who is dealing with the diagnosis orders a MEG scan of the patient’s brain

2. request is sent to instrument which takes a MEG scan and collects data about the activity in the brain

3. This data is then collected and presented to the Grid Resource Broker for analyzing on the Grid

• QOS- deadline and the budget

• optimization method could be one of the three: cost, time or cost-time.

4,5- data and analysis code are dispatched to remote node and results collected

Architecture

Components

parameterization tools (Nimrod-G parameter specification language)

resource broker(Nimrod-G with Gridbus scheduler) grid market directory (Gridbus GMD) low-level grid middleware (Globus) Grid Enabling process: resources - Globus software deployed on them. application - parameter sweep application using the

Nimrod-G parameter specification language. GMD used as a register for publication of resource

providers and services.

Analysis Code developed by the Cybermedia Centre, Osaka University, Japan two phases Phase 1:

• raw data from the brain goes through wavelet transform operation

• time-frequency data of the output Phase 2: cross-correlation analysis is performed for each pair of wavelet

transforms. output displays the similarity between a pair of brain data for

every frequency spectrum.

Wavelet Cross Correlation Analysis

Grid Resource Broker and Scheduler

Resource Broker – Nimrod G + Globus middleware. Gridbus Scheduler. performs resource discovery, selection, and

dispatching of MEG jobs to remote resources. It also starts and manages the execution of jobs

and gathers the results back at the home node. Components of Nimrod G

• A persistent task farming engine.

• A grid explorer for resource discovery.

Grid bus Scheduler

Plugin scheduler- designed to use GMD. Nimrod G- Processing cost based on CPU Time. GMD allows GSP Ao Service+ Service Price. Gridbus Scheduler resource allocation

based on Ao Cost model.

Gridbus Scheduler Algorithms

• Cost minimization.

• Time minimization.

• Cost – Time Optimization. Uses past performance of machines. Average job completion rates.

Grid Market Discovery

Allows service providers to publish services with costs.

Built on standard web service technologies. Client API.

Grid Enabling The Application

Nimrod-G farming engine and dispatcher along with Gridbus scheduler • used for deploying and processing it on Global

Grids

2 programs.• raw2wavelet.

• wavelet2cross.

Meta meg. Time_offset_step.

Pseudo code for meta program

Nimrod –G plan for SPMD.

.$OS $HOME $HOME/alphawave

Plan file for brain activity analysis on the Grid.

Application Deployment and Evaluation

Scheduling Experiments and Result

Deadline= 2 hrs. Budget=1990 Grid $. Summary of Experiment Statistics.

Scheduling with Time Minimization

Scheduling with Cost- optimization

Scheduling with Cost-Time optimization

Visualisation of wavelet analysis results for selected sensors.

Conclusion

The economy based approach of processing brain activity data as illustrated in this paper would help in enforcing QoS requirements of medical applications

Hence would enable adoption of Grid technologies by the bio-instrumentation field.