19
SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University [email protected] Mary Fran Yafchak SURA [email protected] rg Working Group Co-leads:

SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University [email protected] Mary Fran Yafchak SURA [email protected] Working

Embed Size (px)

Citation preview

Page 1: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

SURA GridPlan Infrastructure Working Group

Art VandenbergGeorgia State [email protected]

Mary Fran [email protected]

Working Group Co-leads:

Page 2: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 2/19

Infrastructure WG Charter

Primary Goal:Build a regional grid or

interconnection of semi-regional grids capable of supporting activities that advance the availability of Grid infrastructure to SURA member institutions.

Page 3: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 3/19

Infrastructure WG Charter

Specific Activities:• Increase the number and diversity of

interconnected nodes and their availability to researchers.

• Expand secure grid authentication between institutions, including diverse local CAs & BridgeCA in alignment with national & international directions.

• Implement grid services and tools, and evaluation of these, within a test environment of actual nodes and applications.

Page 4: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 4/19

Infrastructure WG Charter

Specific Activities:• Develop new network monitoring,

measurement and performance tools available to grid applications and users.

• Investigate policy and entry procedures for shared inter-institutional grids & recommend best practices.

• Investigate models to open grid resources to broader user communities (e.g. teaching faculty, students).

Page 5: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 5/19

SURAgrid as Foundation

• Evolved from the NMI Testbed Grid, initiated by SURA in September 2003 as part of the NSF Middleware Initiative (NMI) Integration Testbed Program– http://www.nsf-middleware.org/testbed/testbed_status.asp#grid

• Goal is a scalable infrastructure that leverages local institutional identity and authorization while managing access to shared resources across organizational boundaries. – Working to build the grid without over-specifying the technology – Enabling inter-institutional access that is “seamless” for users

and realistic (interoperable, scalable) for the grid– Fostering the sharing and development of real applications to

proof the infrastructure while bringing recognizable value to users

Page 6: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 6/19

SURAGrid: participants… from May 2004

GSU

UAH

UAB

UMICH

UVA

USC

TACC

Louisiana

TAMU

UFL

OleMiss

GMU

Tulane

UARK

TTU

SC

LSU

GPN

Page 7: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 7/19

SURAgrid Resources SURAGrid Infrastructure ElementsUpdated March 23, 2005

Institution Description OS # node # CPU CPU GHz CPU RAM Grid software Grid sw versionGMU Dell 2 2GMU Sun 3 3GPNGSU in process Linux 24 48LOUISIANALSU Mac G5 Mac OS X 24 48 Globus Toolkit 3.2.1OLEMISS CPU Cluster Pentium III 4 4SC CPU ClusterTACC CPU Cluster Globus Toolkit 3.0.2; 3.2TAMU in process TTU in process 3Tulane AMD Opeteron 2 2 2.0 4UAB Dell Linux 8 32 Globus Toolkit 2.4.3; 3.0.2UAH Linux 4 Globus Toolkit 2.4.3; 3.0.2UArk in process 5 5UFL UMich in process 2 4USC Condor PoolUSC PBS queue linux 4 8UVA Pentium 4 6 6 2.6 Rocks 3.3

TOTAL 91 162

Page 8: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 8/19

Application Example 1

• Genome Alignment Algorithm1. Researcher: Nova Ahmed, PhD Student,

Georgia State University 2. Application Goal: Analysis of genome

alignment performance across clusters and grids

3. Value of sharing: Access to resources not available at home institutions

4. Participating Institutions: GSU, UAB, USC, TACC, UVA

Page 9: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 9/19

Sequence alignment

• Sequences used to find biologically meaningful relationships among organisms

– Evolutionary info; diseases, causes, cures– Finding out information about proteins

• Compute intensive for long sequences– Needleman & Wunsch (1970) - optimal global alignment– Smith & Waterman (1981) - optimal local alignment– Taylor (1987) - multiple alignment by pairwise alignment– BLAST trades optimal results for faster computation

• Challenge - achieve optimal results without sacrificing speed

Page 10: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 10/19

Parallel distribution of multiple sequences

Sequences 1-6

Sequences 7-12

Seq 1-2 Seq 5-6Seq 3-4

Page 11: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 11/19

Computation Time

0

100

200

300

400

500

0 5 10 15 20 25 30

Number of processors

Computation time (sec)

Single Cluster

Single Clustered

Grid

Multi Clustered

Grid

potential for multiple clusters across grid?

Page 12: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 12/19

Run @ UVA using UAB cert (BridgeCA)

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Page 13: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 13/19

Application Example 2

• Task-Farming for Black Hole Simulation1. Researcher: Rion Dooley, IT Analyst, CCT2. Application Goal: Faster calculation for “grand

scale” parameter survey 3. Value of sharing: Unattended, opportunistic use of

computational resources across institutions4. Participating Institutions: LSU, TACC, UVA*, UAB*

*via BridgeCA

Page 14: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 14/19

Task Farming

• Certain classes of problems require large numbers of nearly identical runs to produce meaningful results.

• Examples:– Monte-Carlo simulations– Smith-Waterman analyses– Data mining– Parameter sweeps

• Task farming is a way to utilize multiple resources to solve such problems.

Page 15: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 15/19

Task Farm Infrastructure• Using general Task Farming infrastructure written using

Cactus – Hierarchy of “Task Farm Managers (TFM)”– Pluggable components to easily use different technologies

(e.g. GAT)– Grid enabled and very portable– Supports task scheduling

• Can handle needs of different classes of applications by adding new “Logic Managers”– Fill out simple API for general task farming (how to start

application, provide parameter file, etc)• Application independent

– no need to recompile existing application– Generic Logic Manager can be used for most apps

Page 16: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 16/19

Task Farm Infrastructure

• Grid functionality provided through the Grid Application Toolkit (GAT)– Resource discovery– Job submission– File transfer

• Using GAT means many different technologies/services can be used

http://www.gridlab.org/GAT

Page 17: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 17/19

Task Monitoring

• Task Farming infrastructure can make use of our other tools, e.g.– HTML interface to monitor the

progress of the overall tasks and to steer individual TFM’s*

– Portal interface to start, stop, and track tasks

Page 18: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 18/19

Future of TFI and SURA Grid

• Explore new ways to schedule and share resources with SURA Grid– User-centric vs. resource-centric

resource allocation– Dynamic resource scheduling based

on “good neighbor” policies

Page 19: SURA GridPlan Infrastructure Working Group Art Vandenberg Georgia State University avandenberg@gsu.edu Mary Fran Yafchak SURA maryfran@sura.org Working

March 22, 2005 19/19

Sample Run:Black Hole Simulation Parameter Survey