Computational Grid Using
Commodity Systems
Vijay Potnis
03/31/03
The Concept
Grid -A pool of computing power created using
multiple computers
Goal is to solve a problem that requires enormous
amount of computing power
User submits his work to this pool and gets computed
results
Data Grid - Grid environments tailored to the needs
of the processing of huge data samples
Computational Grid – Grid environment tailored to
the needs of intensive computation
Grid Architecture
How to Build a High-Performance Compute Cluster for the
Grid
Alexander Reinefeld and Volk ErLindenstruth
Framework for high performance cluster that can be used as a
reliable computing node in the grid
Goals

Low total cost of ownership, High Degree of scalability , Maximum
performance for parallel applications
Experiment at European research institution CERN


Large Hadron Collider – high energy physics collider experiment
Huge data generated is reduced by Data selection, filtering &
Compression using this grid
Simple compute farm architecture fails for analysis under quasi realtime condition
…Grid Architecture
Components



Computing nodes – Intel 1.7 Ghz P4, AMD Athlon 1.3 Ghz, Alpha
212644 833 Mhz, with either SDRAM or DDRAM, OS – Linux, PCI(X)
bus
Interconnect & Protocol – To reduce cost Gigabit Ethernet was
chosen. Used Scheduled Transfer Protocol within the cluster and TCP
for communication with outside world
Fault Tolerance – Sensor and actuator function are implemented.
Micro-controller based PCI devices are installed which operated
through a private control network. Repair server take care of other
complex failures such as race conditions.
Integration – Local cluster management system is used to integrate
these individual pieces into a single computational node
Scheduling
Scheduling in a Grid computing environment using
Genetic Algorithm
VincenZo Di Martino and Marco Mililoti
Purpose - Investigate the possibility to use the computing GRID in a
flexible way to permit the maximum usage of resources
Approach - Central repository of resources and resources requests
Two level system of scheduling


First formed by the set of computing nodes – each one with a local
scheduling policy. Local scheduler accepts a single job at a time and allocate
it on the local hardware with respect to the local information
The second level formed by the superscheduler. A superscheduler
(metascheduler) is a centralized scheduler which optimize the allocation of a
job allowing the execution on the fittest set of resources
Developed a simulator to study the usefulness of genetic algorithm for
this scheduler. Simulator is written in C using Parallel Genetic Algorithm
Package (PGAPack) library.
…Scheduling
Given a set of jobs and computing node, superscheduler find
the allocation sequence on each node of a computational grid
that minimize the release time of jobs.
Solution is represented using chromosome encoding a schedule.
For n jobs the chromosome will hold 2* n genes with couples of
two genes representing the job and the node on which it will be
scheduled
Decoding is done by scanning the chromosome and scheduling
each job on the corresponding node with respect to the local
scheduling policy.
Results – for simple configuration and simple problem the
system was able to allocate resource optimally but in complex
configuration the optimum was not achieved.
Developing Applications for a grid
The GrADS Project: Software Support for High-level Grid
Application Development
Francine Berman, Andrew A. Chien, Keith Cooper, Jack Dongarra, Ian Foster, Dennis
Gannon, Lennart Johnsson, Ken Kennedy, Carl Kesselman, John Mellor-Crummey,
Dan Reed, Linda TorcZon, and Rich Wolsky
Goal – To ease process of applications development for grid
Focus:
 Grid software architectures
 Base software technologies, such as scheduling, resource discovery,
and communication
 Programming models, languages, compilers, environments, and
other tools
 Mathematical and data structure libraries, including numerical
methods for adaptivity, control of accuracy, and latency tolerance
Developing Application for Grid
Result - GrADSoft an application framework
It provides an environment where an unmodified parallel application can
be scheduled, instrumented, executed, and monitored in a Grid
environment with no user intervention.
Grid Management
A System for Monitoring and Management of
Computational Grids
Warren Smith
Describes a framework for a System to monitor and manage
computational grids
Work done at NASA on Informational Power Grid which is based on Globus
toolkit.
Framework - Control and Observation is Distributed Environment (CODE)
User creates his own sensors and actuators and add to the base
framework, giving more flexibility to the user in managing any component on
a Grid
This framework is embedded in Globus Resource Allocation Manager
(GRAM) management agent
… Grid management
Security
A Security Architecture for Computational Grids
Ian Foster, Carl Kesselman, Gene Tsudik and Steven Tuecke
Problem - Security concerns when using widespread computational
resources
This paper analyses security requirements of a large-scale distributed
computing and develops security policy and a corresponding security
architecture.
Following are security requirements on a computational grids
 Single sign-on
 Protection of credentials
 Interoperability with local security solutions
 Exportability
 Uniform credentials/certification infrastructure
 Support for secure group communication
 Support for multiple implementations
… Security
Security policies focuses on authentication of users, resources, and processes and
supports user-to-resource, resource-to-user, process-to-resource and process-toprocess authentication
Architecture to Enforce these
Grids using commodity systems
[email protected]
SETI team at UC Berkeley discovered thousands of computers
sitting idle.
Most of them having screen savers and wasting precious
computing resource
This is where [email protected] (and you!) come into the picture.
The [email protected] project uses these free cycles to search out new
life and new civilizations."
On the users acceptance an agent (screen saver) is downloaded.
The screen saver gets a chunk of data over the internet, analyze
that data, and then report the results back to SETI team.
But this is not about using free cycles only
Future Work
Condor Grid Computing from Mobile Handheld Devices
Francisco J. Gonzalez-Castano, Javier Vales-Alonso and Miron Livny
Title is misleading
Goal - Provide user interface to access Condor and not provide
computing power using Mobile Handheld Devices.
But in future you may see SETI running on a PDA or Cell phones
Descargar

Communication between two persons with different …