NGrid Project:

1
NGrid Project
The First Year
Jysoo Lee KISTI Supercomputing Center The
Third PRAGMA Workshop Jan 22-24, 2003
2
2nd ApGrid Workshop at ASCC- NGrid Introduction

• Construction and Operation of the Korean National
  Grid
• NGrid includes
• National Computational Grid
• National Data Grid
• National Access Grid
• National Application Grid (Ex Bio Grid, CFD
  Grid, Meteo Grid etc)
• NGrid is planned at the ministry level by
  MIC(Ministry of Information and Communication)

3
2nd ApGrid Workshop at ASCC Summary

• NGrid Project is a major Grid project!
• Scope everything about constructing a national
  Grid infrastructure
• Budget 32M USD
• Personnel 233 (application), 128 (middleware)
  researchers involved

High hopes for its results!
4
N Supercomputer, Cluster Grid Testbed
5
N Access Grid (including CAVE)
?????? ?????
KREONET KOREN
APII Testbed/KREONet2 STAR TAP
???????
3D(Multi-view) Camera
6
Application Specific Grid TestbedN Bio Grid

• KISTI ??? ??
• Mercury(16), Venus(24), Mars(24),Jupiter(16)
• KAIST ???
• Dual Processor Pentium III
• Single Processor Pentium ?
• ????? ???
• Dual Processor Pentium III
• 4 nodes AMD Cluster
• ????? ???
• Cluster PC 8 nodes
• ????? ???
• Cluster PC 4 nodes
• Cluster PC 8 nodes
• ????? ???

Globus 2.0 MPICH-G2 Linda
7
Application Specific Grid TestbedN CFD Grid

• Condition for analysis
• 2,100,000 unknowns (M 1.7, angle of attack
  90,
• Re 1.431 x 107)

8
NGrid NOC
9
International Cluster Grid
11 institutes 20 supercomputers 8000 CPUS 10
Teraflops
SC2002(Baltimore) 13 countries 41
institutes 3,000 CPUs
10
International Supercomputer Grid
11 institutes 20 supercomputers 8000 CPUS 10
Teraflops
11
NGrid Testbed
Mercury
Venus
Jupiter
Mars
12
NGrid Testbed
kr-mds
gateway / 150.183.249.1
nstarouter / Firewalll
nstargate /
kisti-mds
Ngrid-mds
Gbit Ethernet
mercury node1001
venus node2001
mars node3001
jupiter node4001
Fast Ethernet
Gbit Ethernet
node1002
node2002
node3002
node4002
node2003
node3003
node4003
node1003
node2004
node3004
node4004
node1004
node1015
node2023
node3023
node4015
node2024
node1016
node3024
node4016
16 node
16 node
24 node
24 node
Processor P4 2.0Ghz Mem. DDR 512/node /home
80GB (shared) /scratch 20GB (per node)
Processor P4 2.0Ghz Mem. DDR 512/node /home
80GB (shared) /scratch 20GB (per node)
Processor P4 2.0Ghz Mem. DDR 512/node /home
80GB (shared) /scratch 20GB (per node)
Processor P4 1.7Ghz Mem. RDRAM 1GB/node /home
80GB (shared) /scratch 10GB (per node)
13
NGate Portal of NGrid Project

• Goal of NGate
• Provide user friendly interface
• For novice in Grid environment

14
NGate The Structure

• Web-based interface

MyProxy Server
GSI
JSP Serlvet
Storage Archives
HTML
HTTPS
GSIFTP
Client Browser
HTTPS
ApacheTomcat
Computing Resource
LDAP
Information Service
15
NGate Functions

• Job submission

16
NGate Functions

• MDS Browser
• Provide resource information in tree view
• Secure binding in MDS 2.1

17
NGate Monitoring Service

• Host status information
• Provide host status information registered on MDS
  Server

18
Grid Application Research Areas

• CFD Grid
• High Energy Physics (HEP) Data Grid
• Genomic Grid
• Grid-system for Computational Chemistry and
  Nano-materials
• Nano Materials Computational Grid
• Computing Grid of Comprehensive Environmental
  Flow Model
• Electronic Structure Computation and Informatics
• Development of the High-performance and
  High-Efficiency Heat Exchanger Design Techniques
  Using Parallel-computing
• LES and DNS Methods under Grid Environment
• Realization of Virtual Design Development Grid
  System
• . . .

19
NGrid Middleware Research
20
Some Examples of Grid Application and Middleware
Researches of NGrid Project
21
Demonstration of Full Scale EDG Testbed
CDF CPU
K2K CPU
UI Real user
RB
NFS
In operation
GSIFTP
Big Fat Disk
GDMP client (with new VO)
SNU
SE
VOuser
CE
Disk
LDAP Server _at_SNU
VOuser
NFS
GDMP server (with new VO)
GSIFTP
NFS
GDMP client (with new VO)
GSIFTP
MAP on disk With maximum security
in preparation
NFS
WN
grid-security
GSIFTP
SKKU
VOuser
Disk
KNU
In operation
. . .

• The EDG testbeds are operated at KNU and at SNU
• The globus simple CA is managed at KNU and at SNU
  to sign certs
• In addition to the default VOs in EDG, a cdf VO
  is constructed
• Authentication, VO LDAP, and RC LDAP server is
  running at SNU
• See http//neutrino.snu.ac.kr/bockjoo/EDG_testbed
  /

Center for High Energy PhysicsHEP Data Grid
Working Group
22
(No Transcript)
23
Development of Design Shape Optimization
Technology Using Virtual Wind Tunnel based on Grid
3D Compressible Turbulent Flow Fields Analysis by
Using Efficient Navier-Stokes Solver
Mesh Generation From CAD DATA
Post-processing and Educational Web Environment
Capability of Fast Design Procedure Design
with Large Number of Design Variables
Korea Advanced Institute of Science Technology
24
Realization of Virtual Design
Development Grid System Aerospace Structures
Laboratory Seoul National University

• Constructing VDD(Virtual Design Development)
  GRID System
• Developing IPSAP/G
• (Internet Parallel Structural Analysis
  Program/Grid-enabled) Ver. 1.0
• Carrying out Large-scale simulations ( more than
  100M DOFs )
• Carrying out MDO(Multi-disciplinary Design
  Optimization) of
• 
  aerospace structures in Grid
• Visualization of MDO results and large-scale
  simulations

The cost and time of a design and development
should be minimized to improve the quality of
products. However, verifications through
experiments may be carried out to make a decision
in many cases since the whole real phenomena may
be not simulated accurately in general numerical
analysis and sufficient design parameters may be
not considered. In fact, the most critical and
serious problem is how we can obtain the
cost-effective computing resources for
high-precision simulation and MDO(Multi-disciplina
ry Design Optimizations) fully considering design
parameters. The solution of this problem may be
GRID computing technology. If sufficient
computing resources are provided economically
through GRID computing technology, high-precision
simulations of design results may be carried out
within less than 5 hours and the various
parameters not included before can be considered
in detail. Moreover, the results may be displayed
interactively within design and development
processes and the modifications of design model
can be conducted in cyber space.
Current Research Output

• Large-scale structural analysis
• Platform Multi-cluster
• 2 step mesh-partitioning scheme

Improving parallel performance by considering the
difference of system performance (The ratio of
the number of elements allocated to each
processor reached the system performance ratio)
System Software -OS Linux -Globus (Ver.
1.1.4) -Condor (Ver. 6.3.2) Registered as
globus service -MPICH, PACX-MPI Multi-cluster
communications Application Software -IPSAP/G(Inte
rnet Parallel Structural Analysis Program/
Grid-enabled) -Parallel Multifrontal
Solver(PMS) -Explicit Dynamic Solver -Eigen
Solver(block Lanczos Algorithm PMS) -Nonlinear
Solver (damage, impact), Pre/Post Processing,
Automatic mesh-partitioner -SBEM (boundary
element code for sound simulation) -WFEA
(wavelet packet transformation code for tapping
sound feature extraction)
Three cluster systems (SSC1, SSC2 and SSC3
cluster system) and distributed PC-farm comprise
DCG system. (122 PCs, 170 CPUs)
Cooperative computing
Cooperative computing is a kind of distributed
computing which utilize the distributed resources
to execute jobs simultaneously. In case of
cooperative computing, the communications between
CPUs is not much required since each CPU does
quite independent jobs. Therefore, cooperative
computing may be very suitable for Grid
environment. Cooperative computing may be
utilized for many application filed including
Bio-informatics, parametric study.
The number of jobs allocated into each processor
Grid computing technology was used to develop a
new non-destructive evaluation method. The
computing time for tapping sound calculations was
dramatically reduced through Grid computing
Technology. No additional expense. Grid computing
technology makes TSA (Tapping Sound Analysis) a
feasible and practical methodology S. J. Kim, J.
S. Hwang, and C. S. Lee, "Utilization of
Departmental Computing GRID System for
Development of an Artificial Intelligent Tapping
Inspection Method, Tapping Sound Analysis ",
SC2002 Conference, Technical Paper, Baltimore,
MD, USA, 16-22, Nov. 2002
The elapsed time (Estimated vs Measured)
Seoul Natl Univ.
25
Design Implementation of MPI for GRIDs
Yonsei University
26
Grid Resource Management System

• Research Institution Korea Advanced Institute
  of Science and Technology (KAIST)
• Senior Researcher Seung-Ryoul Maeng
• Research Period Feb. 2002 Jan. 2003
• Research Goal
• Prototyping a grid resource management system
• local resource management system
• resource co-allocator
• Design consideration
• compatibility to our reference model (Globus)
• better performance and scalability
• supports of Linux cluster
• Performance Consideration
• redesign existing GRAM myjob library
• use Glbous_IO library instead of Nexus and DUCT
  library
• redesign existing DUROC control/runtime library
• Local Resource Management System
• Job manager designed as multi-threaded model

KISTI Supercomputing Center
KAIST
27
Development of Grid Middleware Run-Time System (
Fault Tolerance System on MPICH-G2)
Structure

• Project Objects
• Construction of Fault Tolerance MPICH-G2 based
  on Checkpointing Technique
• Feature
• User-Level checkpointing library
• Two phase coordinated checkpointing protocol
• Checkpointing without channel closing
• Guarantee of lost message free
• Failure Detection
• Auto Restart

Job submit with RSL
FaultManager
Gatekeeper
Gatekeeper
MPI application
Job Manager
Job Manager
Wrapper
Ckpt Lib.
MPICH-GF
Process
Process
Operating System
Device Driver
Checkpointing
Recovery

• Testing
• ECDLP (Elliptic Curve Discrete Logarithm Problem)
  based on Polard Rho Algorithm
• Parallel Lattice Reduction Program

Seoul Natl Univ.
28
Thank you! For more information, http//www.gridf
orumkorea.org and http//www.hpcnet.ne.kr