4th PRAGMA Workshop

1
4th PRAGMA Workshop

• June 5-6, 2003
• Melbourne, Australia
• Institutional Update
• by
• Arun Agarwal
• University of Hyderabad
• INDIA

2
Scheme of Presentation

• PART A About the University and recent
• initiatives taken
• PART B Research Application Areas
• PART C Regional Network/GRID activity
• PART D Efforts in Cluster/GRID Computing

3
University of Hyderabad

• Established by an Act of Parliament on October 2,
  1974 as a Central University
• A Unitary Institution, essentially residential in
  character
• Post graduate programmes with strong emphasis on
  research

4
Organizational Structure
Visitor President of India Chancellor
Justice P. N. Bhagwati Chief Rector Governor
of Andhra Pradesh
5
Academic Structure

• School of Mathematics Computer/Information
  Sciences (Mathematics Statistics,
  Computer/Information Sciences)
• School of Physics
• School of Chemistry
• School of Life Sciences (Biochemistry, Plant
  Sciences,Animal Sciences)
• School of Humanities (English, Philosophy, Hindi,
  Telugu, Urdu,
• Applied Linguistics Translation Studies,
  Comparative Literature)
• 6. School of Social Sciences (Economics,
  History, Political Science, Sociology,
  Anthropology, Regional Studies, Folk Culture
  Studies)
• Sarojini Naidu School of Performing Arts, Fine
  Arts Communication
• 8. School of Management Studies

6
Research PerformanceAverage Output per Faculty
1999-00 2000-01 2001-02

• Publications 2.75 2.80 2.79
• Research Projects 0.53 0.58 0.59
• Ph.D. Supervisions 3.80 3.70 3.82
• Ph.D. Awards 0.40 0.53 0.63
• Other Degrees 3.00 2.90 2.80
• Five year average is 1.98 (425 PhD/215 faculty on
  average)
• No. of PhD Theses submitted included
• NOTE Based on average number of faculty

7
Recognition

• one of the five universities in the country
  recognized by the UGC as a University with
  potential for excellence
• accredited by NAAC with five stars
• several departments received special funding from
  UGC, DST, under SAP, DRS, COSIST, FIST
  programmes etc.
• among the 8 Institutions in the High Output
  High Impact Category

8
Awards and Honours-National

• Padma Bhushan
• Padma Shree
• Shanti Swaroop Bhatnagar Award
• INSA Young Scientist Medal
• INSA Fellowship
• INSA Ramanujan Research Professorship
• FNA, FAMS, FNASc, FASc, FNAAS
• FICC, F.Inst.P., C. Phys
• UGC National Fellowship
• UGC National Lectureship
• UGC Career Award

• AICTE Career Award
• Sahitya Akademi Award
• G.D. Birla Award
• Meghnath Saha Award
• Ghurye Award in Sociology
• FICCI Award
• Hari Om Trust Award
• BOYSCAST Fellowship of the DST
• Swarna Jayanti Award
• National Metallurgist Award of Ministry of Steel
• Univ. has provided Directors, Vice-Chancellors
  etc. to other institutions

9
Awards and Honors-International

• UNESCO Travel Grant
• UNESCO Young Scientist Award
• UNESCO ROSTCA Award
• Max Born Award in Quantum Physics
• Fellow of Institute of Physics, London
• Alexander von Humboldt Award Fellowships
• Fellowship of the Third World Academy of Sciences

• Fulbright Fellowships
• Commonwealth Fellowships
• Fellowship of Royal Society of Literature, London
• Fellow of Bunsen Society, Germany
• Editorship of reputed international journals
• Honorary Doctorates from reputed universities
• Third World Academy Award, Trieste

10
InfrastructureRecent Initiatives
11

• Hardware Configuration
•  
•         Central Server (CS) IBM p690
• -          32 Power 4 1.3 GHz Processors
• -          L1/L2 cache
• - 64 GB High Speed Memory
• 540 GB HDD with RAID Control
•   Hardware Management Console with 17
  Color
• Monitor, Keyboard, CDROM, FDD, Mouse
  etc.
• I/O Backup System with the following features
• -          540 GB tape backup
• -          Fiber Channel Adapter 
•       

12

• Hardware Configuration (Contd)
•    
•         System Software
• -          AIX 5.2 OS
• -          Fortran Compiler, C Compiler
  (includes C), etc
• -          Engineering Software/Mathematics/Statis
  tical etc. Subroutine Libraries
• -          Parallelizable compilers
• -          Application Development
  Interface/Environment etc.
•  
•         Graphic Visualization Workstations
• 6 Nos SGI Octane 2 with 2
  R14K Processors, 4GB RAM, 100GB HDD
• 21 Color Monitor, V12
  128MB Graphic Board
• 2 Nos SUN Blade 2000 with 2
  UltaSparc III Cu Processor, 4GB RAM,
• 78 GB HDD, 128 MB Frame
  Buffer, 21 Color Monitor
•  
•        
•  
•       

13
SOFTWARE

• ACCELRYS Suite GCG Wisconsin with SeqMerge,
  Grail Pro Insight II Felix Material Studio
  Cerius 2
• Mathematica, Statistica, Matlab with tool boxes,
  RATS, GAMS, SPSS Version 11.0
• VLSI Simulator (Verilog VHDL) Sythesizer
  Xilinx bundled platform, Placer and Router,
  Waveform Analyzer, FPGA Hardware, etc
• ADF, Gaussian03, Spartan, Mopac2000
• AVID, 3D Studio Max

14
Other Initiatives

• A Proposal is at an advanced stage for
  Establishment of a 1 TeraFlop High Performance
  Computing Facility
• Discussion on setting up of IBM Competence Centre
  in Mobile Computing
• Construction of International Guest House
• Creation of Visiting Faculty programme

15
PART B Our Proposal Building Collaborative
Links
16
Research Application Areas

• Molecular Medicine
• Molecular Modelling
• Molecular Dynamics
• Monte Carlo Simulations
• Drug Design and Delivery
• VLSI Circuit Testing
• Communication Channel Evaluation
• Protocol Design and Validation

• Materials Study
• Economic Development
• Social Empowerment
• Sociological and Historical Data Base Building
  and Mining
• Natural Language Processing
• Marketing Models

17
Projects Requiring High Performance Computing

• Problems in Genomics Bioinformatics
• Biosequencing analysis
• Structural and functional genomics to map protein
  coding regimes and sequencing of DNA
• Wavelet methods to differentiate between normal
  and cancerous tissues
• Mathematical models to relate sequences of amino
  acids to specific details of protein folding

18

• Problems in Molecular Modelling, Quantum
• Chemistry and Computational Drug Design
•  
• Homogeneous and Heterogeneous Catalysis involving
  transition metals using ab initio Quantum
  Chemical Methods
• Combined Molecular Mechanical and Quantum
  Mechanical Approaches in Rational Drug Design -
  Drug Receptor Interactions and chiral
  discrimination
• Design of molecules with large hyperpolarisabiliti
  es for fabricating efficient nonlinear optical
  materials
• Computational Studies to design hard materials to
  replace industrial diamonds
•  

19

• Problems in Electronic structure calculations on
• Molecular Materials
• Molecular Magnetics
• Spin interactions in polyradicals
• Design of Molecular Switches
• Computational Modelling of polymorphs
• Intermolecular interactions in Material Design
• Development of systematic graph-based approaches
  to the analysis of interaction pathways in
  condensed matter

20

• Problems in Physics
• (involving Monte-Carlo and Molecular dynamics
  simulations)
•   
• Large scale phenomena collective fluctuations
  near critical point
• Colloidal emulsions formation of new
  topological defects
• Phase stability of multicomponent mixtures - role
  of molecular structures
• Thin films under antagonistic boundary conditions
  interface phenomena
• Confinement to restrictive geometry - effect on
  phase transitions
• Advanced Monte Carlo (MC) methods - study of
  finite size effects
• Restricted diffusion under confinement - role of
  surface interactions

21

• Problems in Computer Science and Artificial
• Intelligence
• Model the details of language acquisition in the
  mental lexicon and to use this knowledge to
  process natural languages effectively on a
  computer
• Problems in Computer Vision, Pattern Recognition,
  Image Processing particularly, applications such
  as Hand-Eye Robot Control, Stereo Vision,
  Clustering etc.
• Cognitive neuroscience - modeling of learning,
  neural nets, memory, discrimination and
  categorization
• Computational intelligence - quantum computation,
  neural computers, and evolutionary algorithms

22
PART CExisting High Bandwidth Infra-structure

• 1 Gbps Campus Wide Optical Fibre Backbone
• University an ERNET Medium PoP facility 2/8 Mbps
  link to Bangalore, Chennai, Kolkata, which in
  turn is connected to Pune, New Delhi, Mumbai etc

23
(No Transcript)
24
UH Connectivity
Outreach
RF Link
National
VSAT Link
Leased Lines
ISDN Lines
International to be enhanced
Dialup Lines
25
Regional Initiatives

• Proposed participation with CDAC for
  establishment of I-GRID. Each node in the grid
  would have 1 Teraflop processing capability.

26
PART DEfforts in Cluster/Grid Computing
27
Current Projects/Research in CC

• CONDOR Most promising of all.
• University Of Wisconsin, MADISON
• Heterogeneous Environment
• NOW/GLUnix Sun Microsystems, Berkeley Univ, CA.
• Beowolf Linux based cluster SGI
• LSF HPs cluster software
• Load Leveller IBMs PSU USA
• CODINE/DQE etc.,
• All these are originally from CONDOR and
  different variations on basic version of CONDOR

28
Classification of CC

• Based on Focus
• High Performance Challenging Applications
• High Availability Mission Critical Applications
• Based on Ownership
• Dedicated Clusters
• Non-Dedicated Clusters
• Based on Node Architecture
• COPSCluster of PCs, CLUMPS Cluster of SMPS
• COWSCluster of Workstations
• Based on OS Type
• Based on Architecture Homogeneous,Heterogeneous
• Based on Number of nodes

29
Cluster Computing Developments in UH/ADRIN

• Developed a Heterogeneous High Throughput
  Computing Cluster using all the existing
  machines.
• Built a High Availability Clusters for mission
  critical applications using COTS technology.
• Extensions to Grid Technology

30
Comparison with GLUnix and Condor
31

• Centralized Resource Manager Clients
• Daemons At The Client
• Load Monitor (ccd)
• Time to time sends the load state information
  to the scheduler
• Send the job text to the remote node after
  scheduler notifies a match.
• Job Acceptor (Receiver)
• It performs claiming protocol (i.e., checks for
  the requirements and the available resources and
  confirms). Receives the job text.
• Transparent Remote Execution Two necessary
  features of the job exe are its.
• Forks itself to run the remote executable
• Shadow (pgfsd) Perform remote File I/O and send
  it back to the originating machine.
• Daemons At The Server
• Scheduler
• Receive load information from all clients (lmd)
• Score Based Scheduling (sched)

32
Cluster View in UH
Master Node (Cluster Host) Scheduler Web Server
lmd
sched
Node -1
Node -2
Node -3
Node -n
pgfsd
ccd
pgfsd
ccd
pgfsd
ccd
pgfsd
ccd
33
PGFS Design results
Remote Node
Local Node
Application
TCP/IP Or UDP/IP Or RPC  
Shadow Process (setuid process)
Interposition Agent (Library linked with
application)
                             
Standard System Libraries
Standard Library
34
Read
Write
35
Score Based Match Making Framework
  l Mij PiPj? Wjk Nik (1-Li)
k1
Score based Scheduling or Matching 2 (Master
Node)  
Match Notification 3
Match Notification 3
Advertize 1 resources
Resource Requestor (Remote node where job is
submitted)  
Resource Provider (Node owner)
Claiming 4  
n,m Schedule at T MAX ?
Mij i,j1
36
Our Load Sharing/Load balancing Approach

• Our Approach Condor GLUnix
• Centralized -do- Centralized
• and hierarchical
• Adaptive -do- -do-
• Server Side Bidding NA NA
• Sender Initiated -do -do-
• Pre-emptive -do- -do-
• Node state based -do- -do-
• Group Scheduling Single Single
• Best Combination Best Fit Best Fit

37
High Availability Cluster using RS-232 link
Private Network RS-232
Heart Beat
Storage
Public Network
Client
Client
Client
Client
38
CLUSTER HOME PAGE

• Resource Providers (owners) Resource
  Requestors(users) access this page for their Ads.

39
Adding A Node into the CLUSTER

• Owner of the node make his/her node available
• by submitting this form to the master node.
• Details denote processor, OS, IP address,n/w,RAM
  etc

40
Owner Constraint Form

• Owner of the node specifies the conditions under
  which
• his node will be available to other users.
• User can specify the priorities to the users,
  groups.

41

• Job Submission details are provided by the
  requestor
• Job details like the local directory, RAM, Disk
  Space, estimated time etc.
• Requestor could also provide details about no.
  processors, OS required etc
• Job specs like float operations, integer, IO at
  macro level can optionally be given

42
Existing Cluster Configuration in ADRIN
SGI Origin 2000 2X195 MHz R10K Processor 256 MB
RAM 40 GB SCSI FW Disk IRIX64 Ver 6.4 (Master
Node)
100 Mbps Network
HP 1X99MHzPA RISC 64MB RAM 2 GB SCSI-II 10 Mbps
NIC HP-UX 10.02
Sun Ultra Ultra Sparc-1 167 Mhz 128 MB RAM 40GB
FW SCSI Solaris 2.6 O.S
Sun Ultra-80 4X450 MHz Ultra Sparc-II 2GB
RAM 2X40GB Ultra SCSI Solaris 2.7 O.S
Sun Ultra 1X167 MHz Ultra Sparc-I 128 MB RAM 40GB
FW SCSI Solaris 2.6 O.S
43
Future direction of Work

• Grid scheduler design implementation
• Integrate PGFS with PVM and MPI
• Compare the results of scheduler design with
  other
• Augment Windows NT/XP OS to SSI layer

44
Developments towards GRID

• Extending the philosophy of Score Based
  Scheduler to design a scheduler, which will aim
  at maximizing the Cluster utilization in UH
• Extend the interfaces to support submission of
  jobs over the internet.
• Include security features to protect the internal
  information and also maintain complete
  confidentiality of the submitted jobs.

45
Efforts in Bioinformatics
46
Our Mission

• the theoretical modeling of the living
  processes with the final objective of application
  or manipulation of living system for betterment
  of the quality of human life

47
Objectives

• pure theoretical research genomics and
  proteomics and related topics. Studies leading
  to discovery of new genes etc. Protein structure
  and function studies. Interrelation and
  interdependence of cellular subsystems
• semi-applied research Diseases and cures.
  Preventive medicines. Genetic defects. Other
  related research
• applied research Drug discovery and drug design.
  Development of tools.
• development of bioinformatics tools (database
  management, intelligent tools, new algorithms for
  searching etc.)

• software designs to use the exiting data for
  some given application.
• new tools of verification of algorithms This is
  an area that needs lots of attention as we must
  ensure that the software does what it is supposed
  to do and there is no side-effects
• Biocomputing
• development of teaching material We need new
  people to continue old knowledge and add new
  ideas. (new ideas come from the young generation)
• information sharing (e.g., development of web
  based resources rice genomics).

48
Large Scale Gene Sequencing Programs Japan-
Genomic and Gene sequences, as a part of
International Rice Genome Sequencing Program of
which India is a member , Univ Delhi and
IARI Internatl Rice Research Inst., Philippines
Functional Genomics and sequencing USA Gene
Sequencing for salt tolerance ChinaGenomic and
Gene sequences India- University of Hyderabad
Gene sequencing and Functional genomics (eg
Drought tolerance in Rice). Funded by
Rockefeller Foundation, USA To date UH has
contributed the most sequence information for the
expressed genes of rice from India
49

• University of Hyderabad Rice Functional
• Genomics Program
• Search for genes that control water stress
  response process in rice, Experimental approach
  via making gene libraries
• DNA Sequence Characterization of such
  genes/clones using advanced automated DNA
  sequencing facility.
• Functional definition of the role of these genes
  in drought tolerance.
• It is critical to identify genes in rice genome
  that are associated with drought tolerance in
  order to deploy them by rice breeders.

50
Automated DNA Sequencing Facility (MegaBace 500)
Plant Molecular Genetics Rice Functional
Genomics Laboratory University of Hyderabad

• Total number of clones sequenced 15,000
• Sequencing for EST generation Gene
  Discovery 10,000
• Number of Expressed Rice Genes Generated and
• deposited in GenBank as on April 30, 2003
  6,000
• GenBank Accession Nos BI305180 to BI306756,
• BU672765 to BU306756,
• CB964418 to CB967504.

51
THANK YOU