Transcript Grid Computing – Introduction

Grid Computing –
Introduction
Sathish Vadhiyar
Generic Grid
Architecture/Components
Grid Access & Info
Problem Solving
Environments
User Portals
Service Layers
Authentication
Scheduling &
Co- Scheduling
Computers
Naming &
Files
Application Science
Portals
Resource Discovery
& Allocation
Events
Data bases Online instruments
Resource Layer
High speed networks and routers
Fault Tolerance
Software
OK, I have built some software.
Is mine a Grid software?
Ian Foster’s three-point checklist:
1. coordinates resources not subject to
centralized control
2. using standard, open, general-purpose
protocols and interfaces
3. to deliver non-trivial qualities of
service
Some Myriad Definitions
 “Coordinated resource sharing and problem
solving in dynamic, multi-institutional virtual
organizations”
 “Anatomy of the grid – highly flexible sharing relationships,
sophisticated and precise levels of control over use of shared
resources, sharing of varied resources, diverse usage modes.”
 “Controlled sharing – not free
access”
 “Infrastructure enabling integrated, collaborative use

of resources”
“Sharing resources can vary dynamically vary over time”
 More colorful definitions keep coming
 Common keywords: Coordinated, shared, multi-institutions,
controlled, usage, collaboration
Differences with Other
Technologies
 Enterprise-level distributed computing – limited




cross-organizational support
Current distributed computing approaches do not
provide a general resource-sharing framework that
addresses Virtual Organization (VO) requirements.
WWW – just client-server. Lacks richer interaction
models
Technologies like CORBA, Java, DCOM – single
organization, limited scope
Some of the Grid techniques complement existing
techniques.
Grids vs Conventional Distributed
Computing (Nemeth and Sunderam)
 Distributed Computing
 Virtual Pool of nodes
 Set of nodes static. Users have login access. They explicitly know
about nodes
 VM constructed out of a priori knowledge
 Resource assignment implicit
 Resource owning
 Grid Computing
 Virtual Pool of wide range of resources
 Set of nodes static/dynamic. Resources dynamic and diverse – can
vary in number, can vary in performance
 Difficult for user to get a priori knowledge
 User abstraction at resource layers
 Resource sharing
 Apps. – resource requirements more than can be solved on machines
“owned”
Continued
Nemeth and Sunderam
Motivating examples
SETI@home
 To search new life and civilizations
 Use individual computers’ idle time through
running SETI@home screen saver
 Screen savers retrieves data, analyzes and
reports results back to SETI project
 Looking for extra-terrestrial signal over a 12second period
 Each work unit takes 10 to 50 hours on an
average computer – 2.4 to 3.8 trillion floating
point operations
Steps and Statistics
Data collected from Arecibo telescope in Puerto Rico onto
tapes and shipped to SETI@home lab in UC, Berkeley.
Break tapes -> work units -> given to users
Find candidate signals reported from users
Other steps:
•Checking data integrity
Statistics
from 1999-2004
•Removing radio frequency interference
(RFI)
Users
•Identify final candidates
Results received
Statistics:
Total CPU time
208,174,383 work units
Floating Point
1,261 tapes
Operations
Images and statistics from SETI
web site
Total
5054812
1459999962
1988719.151 years
5.278185e+21
Average CPU time
11 hr 55 min 56.3 sec
per work unit
Climateprediction.net
 Forecast climate in 21st century
 3 steps – explore current model,
validate against past climate,
forecast 21st century climate
 Different models (in terms of
initial conditions, forcing
[volcanoes, solar activity etc.],
parameters [approximations or
ranges of fixed values in the
model. E.g. ice size in ocean,
friction between different ocean
layers]) distributed to different
users
 Massive ensemble experiment
From
climateprediction.net
Steps
Experiment
1
Explore model
sensitivity to
parameters
Goal
Identify suitable ranges
of parameters.
2
Simulation of 19502000
Assess model skill by
making a probability
based forecast of the
past climate.
3
Simulation of 20002100
Make a probability based
forecast of future
climate.
From climateprediction.net
Methodology
Each simulation includes 3
phases:
• Calibration (15yrs)
• Pre-industrial CO2 run
(15yrs)
• Double CO2 run (15yrs)
Run the model with a range of
initial conditions and
parameters for the period
1950-2000. Compare model
outputs with observations to
assess how well the model
performs.
Run the model with a range of
initial conditions, forcings
and parameters for the
period 2000-2100.
Prime number generation - GIMPS
 Finding Mersenne prime numbers – 2P-1
 GIMPS is to find largest known Mersenne





prime numbers
41st Mersenne prime found recently 224,036,583-1 with 7,235,733 decimal digits !!!
GIMPS found seven
For mostly fun
1000s of Pentium PCs involved. Setup similar
to SETI@home
PCs do primality tests
Other @home Projects
 genome@home – designing new genes that form






working proteins in cells. To study protein evolution.
Individual PCs design protein sequences
folding@home – to study why proteins fold/misfold.
Each PC simulates a particular kind of protein folding
evolution@home – to understand and simulate
evolution
Compute-against-cancer – to study cancer cells and to
design new cancer drugs
FightAids@home – screen millions of candidate drug
compounds
Distributed.net – cryptography, secret key challenges
More can be found in
http://boinc.berkeley.edu/projects.php
The Telescience project
 Grid for remote
accessing microscopes,
data analysis and
visualization
 To study complex
interactions of
molecular and cellular
biological structures and
hence understand brain
diseases
 Interactively steer a
400,000-volt electron
microscope at UC San
Diego
From TeleScience web site
References
 http://www.globus.org/research/papers/chapter2.pdf
 What is the Grid? A three point checklist. Ian Foster. GRIDToday,





July 20, 2002.
The Anatomy of the Grid: Enabling scalable virtual organizations. I.
Foster, C. Kesselman, S. Tuecke. IJSA. 15(3), 2001.
A Complete History of the Grid. Dr. Rob Baxter. Pdf
Zsolt Nemeth, Mauro Migliardi, Dawid Kurzyniec and Vaidy
Sunderam. A comparative analysis of PVM/MPI and computational
grids. In EuroPVM/MPI 2002.
Zsolt Nemeth and Vaidy Sunderam. A comparison of conventional
distributed computing environments and computational grids. ICCS
2002.
Zsolt Nemeth and Vaidy Sunderam. A formal framework for defining
grid systems. CCGrid 2002.