MCell Usage Scenario Project 7 CSE 260 UCSD

MCell Usage Scenario Project #7 CSE 260 UCSD Nadya Williams nwilliams@ucsd. edu

Outline • What is MCell ? • How to run MCell ? • Resources • Usage Scenario • Summary

What is MCell ? A General Monte Carlo Simulator of Cellular Microphysiology … MCell now makes it possible to incorporate high resolution ultrastructure into models of ligand diffusion and signaling … Thomas M. Bartol Jr. Joel R. Stiles Computational Neurobiology Laboratory Neurobiology & Behavior Cornell University The Salk Institute

What is MCell ?

What is MCell ? MCell uses • Monte Carlo diffusion • Chemical reaction algorithms in 3 D MCell simulates • Release of ligands in solution • Creation/destruction of ligands • Ligand diffusion within spaces • Chemical reactions undergone by ligand effector

What is MCell ?

What is MCell ?

What is MCell ? Main biochemical interactions • 3 D diffusion of ligand molecules based on Brownian motion • the average net flux from one region of space to another depends on molecules mobility depends on 3 D concentration gradient between the regions

What is MCell ? Different approaches to computing 3 D gradients Monte Carlo approach With Voxels Assume well-mixed condition Use PDEs for average net changes • correct behavior PROS: average system CONS: • too complex for realistic structures • output has no direct stochastic information • Directly approximate the Brownian movements of the individual ligand • Chemical reaction rates are solution rate const PROS: • events are considered on a molecule-by-molecule basis • the simulation results include realistic stochastic noise CONS: • complexity

How to run MCell ? Simulate the system behavior • Running the same computation with different seeds • Averaging all the instances Input Data consists of • one or more MDL scripts • files describing elements of the simulation spatial geometry effector location chemicals' repartitions Each instance has • A pre-defined number of time steps • Input data Output files • resulting stochastic model • visualization files

Resources Typical run now: Run envisioned: • 5 MBytes of input data per task • 1000 tasks • 1 MBytes 2 -D output files per task • 10 MBytes 3 -D output files per task • usually 100 MBytes of RAM • require on the order of 10 minutes of processing on today's most powerful CPUs. • Modeling ligands exchange, diffusion • 50 MBytes of input data per task • 1, 000 tasks • Tens of GBytes 2 -D and 3 -D output files per task • RAM not easily available to an average user • CPUs of MPPs. • Modeling entire cells

Resources Salk Institute UCSD U. of Tennessee Bartol and Sejnowski Casanova and Berman Dongarra and Wolski MCell executes multiple instances of a given code on different parameter set and collects (and perhaps processes) the results. PROS: each instance is independent from the others each instance can be executed anywhere Challenges: 1 tasks share common files 2 resource discovery 3 fault detection 4 fault recovery 5 scheduling

Usage Scenario

Usage Scenario • • • Security Requirements data confidentiality need for digital signatures, encryption, authorization public vs. private information on application status and execution Performance Requirements network bandwidth latency and jitter CPU load information service query time disk capacity, speed application timing formats

Usage Scenario Programming Model • user interfaces (submit, monitor, steer runs) • support for data analysis and visualization Information Service Requirements • frequency of information access • application preferences on location, structure, • representation, and format of IS information : CPU RAM Disk Network Queue waiting time

Usage Scenario • • • Scheduling Requirements resource reservation application components, computation data, intermediate files remote instruments tolerance to delays during execution Remote Data Access requirements publication, management, storage streaming vs. batch processing User Services system status, its format application needs for system services and tools

Summary The MCell development contributions: • larger problem size model for a class of science applications • parameter sweep application model for the Grid. MCell needs: • large-scale MCell runs • further improvement and development of application scheduling mechanism

Milestones • • What are current problems and bottlenecks ? Can one improve basic usage scenario ? Current needs of application from GIS What are requirements for – job scheduling, – job control – storage infrastructure