Biomolecular Path Sampling Enabled by Processing in Network Storage

1
Biomolecular Path SamplingEnabled by Processing
in Network Storage
P. Brenner, J. Wozniak, D. Thain, A. Striegel,
J. Peng, and J. Izaguirre

• Department of Computer Science and Engineering
• Department of Chemistry and BioChemistry
• University of Notre Dame

2
Research Motivation

• Scientific discovery and optimal design through
  shared heterogeneous grid resources
• Collaborative computing promotes max utilization
• Shared capital expenditures

• Computational biochemistry
• Conformation Sampling
• Committer Probabilities
• Rate Calculations
• Docking

3
Our Grid Environment

• Heterogeneous
• Operating Systems, Architecture, Software
• Autonomous
• Frequent random evictions by owner/user
• Both computation and storage
• Incremental check pointing (not dynamic)
• Testbed snapshot (www.nd.edu/ccl)
• Computation 468 CPUS
• Storage 31.3 TB
• Bandwidth 10Mb, 100Mb, 1000Mb

4
The PINS Framework

• Condor matchmaking
• Locality aware storage (file generation,
  replication, access)
• Minimal client side bandwidth consumption

5
The GEMS Framework

• Shared storage fabric built upon Chirp
• Abstracted locality with hybrid database
  interface

6

• Shared

7
Robust

• Server upgrade all jobs expired

• Re-ran submission script automatic continuation

8
High resolution data generation

• In-network bandwidth utilization
• Data generated and stored in place
• Prioritized replication
• Collocation weighted post processing
• Locality aware remote access capability
• Biomolecular Example
• 100 trial trajectory paths to capture committer
  probabilities for side chain (dihedral) motion
• Each high resolution trajectory is 3GB

9
Preliminary WWd Paths

• Explicitly solvated NVT trajectory
• The Order Parameter
• ARG12 backbone dihedral
• Investigating correlations of side chain motion
  to NMR
• Other possible contributions
• Transition Path Sampling
• Chandler, Dellago, Bolhuis
• Committers/Transmission
• Snow, Rhee, and Pande (2006)

10
Path Points
Starting Dihedral Value

• 50 trajectories from 2 trial points

11
Tradeoffs and Challenges

• Efficiency
• Total computation time can be much less than the
  observed completion time
• Large variation in CPUs (utilization vs speed)
• Large variation in resource evictions
• Checkpointing frequency
• Computation efficiency is proportional
• PINS/GEMS overhead inversely proportional
• Additional scripting for fault tolerance
• Automatic continuation is not currently an
  application generic framework feature

12
Acknowledgements

• Advisor
• Dr. Jesús Izaguirre
• Funding Agency
• NSF Grant DBI-0450067
• Collaborators
• Dist Sys Dr. Striegel, Dr. Thain, Mr. Wozniak
• Biochemistry Dr. Peng
• Website
• http//gipse.cse.nd.edu/GEMS/

13

• Questions