MGrid: A Molecular Simulation Grid on the KGrid Testbed

1
MGrid A Molecular Simulation Gridon the KGrid
Testbed

• Karpjoo Jeong
• Bio-Molecular Informatics Center
• Konkuk University

2
Outline

• Objective
• Research Scope
• Execution Strategy and Plan
• MGrid System Overview
• Progress Summary
• Further Research

3
Research Objective
????

• Motivation
• Need for Grid Computing
• Molecular simulation-based applications have
  challenging computation requirements
• Grid computing is a promising approach
• Obstacles
• The scientist has limited knowledge about grid
  computing technology
• Molecular simulation-based applications have
  specific functional requirements (e.g.,
  interactive monitoring) that conventional grid
  middleware fails to address effectively
• Goal
• Develop a user-friendly application grid system
  to be customized for large scale molecular
  simulation-based experiments on top of general
  purpose cost-effective large scale grid computing
  system
• Carry out real world BT/NT applications on MGrid

4
Mutually-Conflicting Motivations
General Purpose (Efficient) Grid Computing System
Challenging Computation Requirements
BT/NT Application
Application Specific Requirements
Application- specific Grid System
Motivations for MGrid
5
Research Scope

• System
• Build MGrid (a molecular simulation grid system)
  on top of KGrid testbed
• Problem Solving Environment (PSE). Kookmin Univ.
• Distributed Job Server
• Grid Portal
• Operate MGrid on KGrid Testbed Production Grid
• BT/NT Applications
• Build a chiral separation database
• Carry out a few large BT experiments

6
Research Execution Strategy and Plan
PRAGMA GGF
Third Party Institutes
? System Support
? System Demo
? System Evaluation
(IT) Grid System Development
? System Evaluation
MGrid System Development
MGrid System Operation on KGrid Testbed
? System Evaluation
? functional Requirements
? System Support
? Requirement Formalization
7
Introduction to MGrid (Molecular Grid)

• Application Grid System
• Designed for large scale molecular
  simulation-based bio/nano research
• Running as application system on top of the
  NGrid testbed which includes clusters and
  supercomputers from several institutes in Korea
• Globus-based Computational and Data Grid System
• Problem Solving Environment with
  Application-specific support and Grid-transparent
  Design
• Distributed Job System with Grid-facilitating
  Design
• Grid Portal with Grid-centralizing Design
• A Spin-off Project from KGrid Project at KISTI

8
MGrid on KGrid Testbed
Supercomputer
Cluster

• KISTI
• - CPU POWER 4 8
• - RAM 16 GB
• - HDD 500 GB
• CBNU
• - CPU PowerPC 32
• - RAM 24 GB
• - HDD 726 GB
• MYUNJI
• CPU R12000 16
• RAM 25GB
• HDD 90GB 16
• TIT
• - CPU PowerPC 54
• POSTECH
• - CPU PA 8500 16
• - RAM 32 GB
• - HDD 288 GB

• KISTI
• - 80 nodes
• POSTECH
• - 40 nodes
• PNU
• - 13 nodes
• UOS
• - 16 nodes
• SNU
• - 16 nodes
• KKU
• - 30 nodes
• KMU
• - 16 nodes

SNU, UOS, MJU, KKU, KMU
KISTI
POSTECH
CBNU
TIT,PNU
9
Collaborators

• Konkuk University
• Karpjoo Jeong (jeongk_at_konkuk.ac.kr)
• Seunho Jung (shjung_at_konkuk.ac.kr)
• Kookmin University
• Suntae Hwang(sthwang_at_kookmin.ac.kr)
• KISTI (Korea Institute of Science and Technology
  Information)
• Jysoo Lee (jysoo_at_kisti.re.kr)
• Kum Won Cho (ckw_at_kisti.re.kr)
• Gee Bum Koo (voxel_at_kisti.re.kr)

10
MGrid Structure
PSE
PSE
Grid Portal
XML interface
Distributed Job Server
Distributed Job Server
Distributed Job Server
11
MGrid Design Focus
MGrid
12
Problem Solving Environment (PSE)
Grid-transparent Design

• A Unicore style of functionality and GUI
• Grid-transparent Design
• GUI-based Single System View Support
• Distributed Repository System for Simulation
  Results
• Fault-tolerant Meta-scheduling
• Application-specific support
• Support for Simulation-based Experiment Planning
  in a Workflow Style
• Interactive Real Time Simulation Control
• User-defined Event Notification
• Visualization tool plug-in support
• Uniform support for legacy simulation software

13
Problem Solving Environment (PSE)
14
Distributed Job Server Grid-facilitating Design

• Object-oriented Design
• Simulation job as an object
• Object-oriented software design
• Open service design
• XML-based open interface via GRAM
• Uniform interface to legacy software
• Job Execution Framework
• Uniform access support for simulation results
• URL-based naming
• GridFTP-based global access
• Local authority-respecting design
• Dispatch tasks via the local resource manager

15
Distributed Job Server
PSE
Grid Portal
Globus GASS
XML-based Interface
Globus GRAM
Globus GridFTP
Job Manager
Event Manager
Shared Repository
Shared Event Queue
Local Resource Manager (e.g., PBS, Condor)
Job Execution Framework
Job Execution Framework
Gaussian98 Driver
Gaussian98 Driver
GAMESS Driver
GAMESS Driver
CHARMM Driver
CHARMM Driver
Distributed Job Server
16
Grid Portal Grid-centralizing Design

• Account Management
• Monitoring Service
• Job/Task Progress Monitoring
• Support search by job names and description
• Provide info about simulation result files
• Resource Monitoring
• Response Time
• CPU load
• Number of current jobs
• Job Management Service
• Submission and Termination
• Grid Administration Service

17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
Real World Application Chiral Separation Database

• Analyzing chiral drug candidates by docking them
  with chiral selectors
• Chiral drug candidates (guest) 1000 for now
• Chiral selectors (host) 50 for now
• Chiral Separation Database and Host Prediction
  Method
• So far, real experiments have been mostly used,
  but take a couple of years for a single pair of
  guest and host. Selecting a right host is very
  important
• By building and analyzing a database about host
  and guest docking, develop a host prediction
  method
• High Throughput Grid Computing
• For a single workstation, molecular simulation
  for a single pair of guest and host takes about
  two weeks
• Molecular simulation for 100050 pairs takes
  2,000years with a single workstation. With high
  throughput grid computing, we can shorten this
  time significantly
• Current Status 200 experiments are currently
  being executed

24
Experiment Chiral separation by
Cyclocarbohydrates

Enantioselectivity MC simulations
25
Progress Summary

• System
• MGrid software system (version 0.9) is
  implemented
• MGrid has been deployed on KGrid Testebd
• Application Research
• 200 Chiral Separation Experiments are currently
  being executed
• Publication
• Application Area
• five SCI papers
• two international conference papers
• System Area
• two SCIE papers
• Presentation at Forum
• Presentation at PRAGMA 4 and 5
• System Demo
• PRAGMA 5
• SC2003

26
Further Research

• Standardize the administration of KGrid Testbed
• MGrid Grid Portal can automatically figure out
  the configuration and monitor the KGrid testbed
  in a real time manner
• Develop a distributed monitoring system for
  KGrid testbed
• Application-level, Globus-level, and
  Network-level monitoring
• Upgrade KGrid Testbed to be a production grid by
  the experience with MGrid
• Develop the OGSA version of MGrid