High Performance Computing Center Texas Tech University

1
High Performance Computing CenterTexas Tech
University

• Phil Smith
• Director
• www.hpcc.ttu.edu

2
Multivariate Minimization using Grid Computing

• Subtitle Recruiting Grid Users
• Co-authors
• Kandle Kuhlish Grad Student
• Jerry Perez -- Grid Administrator

3
Recruit Users

• Typical (interesting) Example
• Easy to understand
• Easy to run
• Demonstrates grid features
• Parallel execution
• Discovery of resources
• Scheduling

4
Global Optimization

• Min f(x) x in ? ? Rn
• Where possible, use the grid to overwhelm this
  problem
• Random search
• Simulated annealing
• Sophisticated descent methods
• Domain decomposition

5
Objective Function

• Implicit ODE BVP
• F(x, y, y, y) 0 y(0) y(1) 0
• Finite Difference approximation
• xj j/(n1) j 0, , n1 1/h
• Solve for yj satisfying F(xj, yj, ? yj
  /2h, ?2 yj /h2) 0
• min max j F(xj, yj, ? yj /2h, ?2 yj /h2)
  all yj

6
TechGrid

• Over 150 windows PCs
• Pleione (Origin2000)
• Antaeus, Mathwulf, Weland
• AVAKI middleware
• MPICH for MPI jobs

7
Student Lab
8
Weland
9
(No Transcript)
10
Grid Submission Modes

• Single job submission
• avaki -v run --execmini_no_p
  -Sspecfile -ssched
• Multi-run One .exe multiple files
• avaki -v run --execmini_no_p -n3
  -Sspecfile -ssched
• MPI
• avaki -v run --execmini_p qtasks_per_node4
• h/hosts/ATLC-MPI-frontend inpwdfile
  inmachinefile stdoutoutput stderroutput
  outoutput

11
Sched Files

• /hosts/EHM039 1
• /hosts/EHM072 1
• /hosts/EHM077 1

12
Spec file

• IN input.txt c/av/mini_no_p/input.txt
  
• OUT output.txt c/av/mini_no_p/output.txt
• Or if you want to store your results in grid
  space
• IN input.txt c/input.txt
• OUT output.txt avaki/home/vij/output.txt

13
Input Files

• input1.txt
• 3 300000
• 1.0 1.0 1.0
• 7.0 7.0 4.0
• input2.txt
• 3 300000
• 1.0 1.0 4.0
• 7.0 7.0 7.0

14
Typical Output

• output1.txt
• 3
• 300000
• 1.000000 1.000000 1.000000
• 7.000000 7.000000 4.000000
• Min found 3.042426
• Where 1.044611 1.021356 1.047931

15
Challenges

• Develop best practices for grid users
• Naïve users versus advanced users
• Working in a multi-level and multi-resource
  heterogeneous environment
• Education in parallel computing
• Debugging in a grid environment is an order of
  magnitude harder than in a stand alone setting.

16
Admin Challenges I

• We do not own most of the Grid resources
• Grid zone admins do not work for me
• Different management styles
• Different work ethic and motivations
• Grid machines are spatially dispersed
• Debugging issues
• Students/users turn the machines on and off
• Power failures

17
Admin Challenges II

• Manpower
• Grids are manpower intensive
• User education
• Administration issues
• Organizational issues
• One Grid admin and 2 (1/2 time) RAs

18
Small user base

• CS grid researchers
• Math and CS
• Cryptography (Factoring large integers, Pollards
  rho)
• PDEs and Stochastic ODEs
• Optimization
• Chemistry
• Statistics of atomic trajectories
• Architecture
• Render Farm

19
Develop Grid Examples

• Address interesting problems
• Easy to understand
• Utilize Grid Resources
• Scientific Grid Services
• Optimization
• Simulation
• Rendering