Excitable Cell Simulation In Star-P

Slides:



Advertisements
Similar presentations
Pinpointing the security boundary in high-dimensional spaces using importance sampling Simon Tindemans, Ioannis Konstantelos, Goran Strbac Risk and Reliability.
Advertisements

TU/e Processor Design 5Z032 1 Processor Design 5Z032 The role of Performance Henk Corporaal Eindhoven University of Technology 2009.
11Sahalu JunaiduICS 573: High Performance Computing5.1 Analytical Modeling of Parallel Programs Sources of Overhead in Parallel Programs Performance Metrics.
Data Analytics and Dynamic Languages Lee E. Edlefsen, Ph.D. VP of Engineering 1.
CS 584 Lecture 11 l Assignment? l Paper Schedule –10 Students –5 Days –Look at the schedule and me your preference. Quickly.
Copyright © 1998 Wanda Kunkle Computer Organization 1 Chapter 2.1 Introduction.
10 minutes overview to z-tree. Outline of Presentation What is z-tree? What can it do? How does it work? When to use? Alternatives? A quick demo.
Linear Simultaneous Equations
Yavor Todorov. Introduction How it works OS level checkpointing Application level checkpointing CPR for parallel programing CPR functionality References.
GPU-accelerated Evaluation Platform for High Fidelity Networking Modeling 11 December 2007 Alex Donkers Joost Schutte.
Matthew Fischels Aerospace Engineering Department Major Professor : Dr. R. Ganesh Rajagopalan REDUCING RUNTIME OF WIND TURBINE SIMULATION Los Alamos National.
SPMD: Single Program Multiple Data Streams
Scheduling Many-Body Short Range MD Simulations on a Cluster of Workstations and Custom VLSI Hardware Sumanth J.V, David R. Swanson and Hong Jiang University.
1 Challenge the future High accuracy machines on factory floors Anthonie Boogaard.
April 26, CSE8380 Parallel and Distributed Processing Presentation Hong Yue Department of Computer Science & Engineering Southern Methodist University.
Matlab Demo #1 ODE-solver with parameters. Summary Here we will – Modify a simple matlab script in order to split the tasks to be sent to the cluster.
Scaling Area Under a Curve. Why do parallelism? Speedup – solve a problem faster. Accuracy – solve a problem better. Scaling – solve a bigger problem.
Building a Real Workflow Thursday morning, 9:00 am Lauren Michael Research Computing Facilitator University of Wisconsin - Madison.
Group May Bryan McCoy Kinit Patel Tyson Williams.
Copyright © 2011 Curt Hill MIMD Multiple Instructions Multiple Data.
Making Watson Fast Daniel Brown HON111. Need for Watson to be fast to play Jeopardy successfully – All computations have to be done in a few seconds –
Parallel Computing With High Performance Computing Clusters (HPCs) By Jeremy Cathey.
Faucets Queuing System Presented by, Sameer Kumar.
COSC 2007 Data Structures II Chapter 13 Advanced Implementation of Tables IV.
Distributed simulation with MPI in ns-3 Joshua Pelkey and Dr. George Riley Wns3 March 25, 2011.
Matlab for Engineers Gari Clifford © Centre for Doctoral Training in Healthcare Innovation Institute of Biomedical Engineering Department of.
EGEE-III INFSO-RI Enabling Grids for E-sciencE EGEE and gLite are registered trademarks Enabling the use of e-Infrastructures with.
HPC HPC-5 Systems Integration High Performance Computing 1 Application Resilience: Making Progress in Spite of Failure Nathan A. DeBardeleben and John.
Scaling Conway’s Game of Life. Why do parallelism? Speedup – solve a problem faster. Accuracy – solve a problem better. Scaling – solve a bigger problem.
STDP and Network Architectures Parallel ODE Solver and Event Detector Eugene Lubenov.
Kia Manoochehri.  Data Analysis  Learn about AWS Struggle to do so  Data Visualization  An exercise in scalability.
1 Potential for Parallel Computation Chapter 2 – Part 2 Jordan & Alaghband.
Introduction. News you can use Hardware –Multicore chips (2009: mostly 2 cores and 4 cores, but doubling) (cores=processors) –Servers (often.
Supercomputing in Plain English Tuning Blue Waters Undergraduate Petascale Education Program May 29 – June
Lecture 3 – MapReduce: Implementation CSE 490h – Introduction to Distributed Computing, Spring 2009 Except as otherwise noted, the content of this presentation.
INTEGRATOR CLASSIFICATION & CHARACTERISTICS
Differential Equations and a few helpful ways to solve them
Parallel Computing and Parallel Computers
The SCEC CSEP TESTING Center Operations Review
Lecture 2: Performance Evaluation
Virtual Machines What is a virtual machine?
September 2 Performance Read 3.1 through 3.4 for Tuesday
Introduction Super-computing Tuesday
Introduction to parallel programming
ECRG High-Performance Computing Seminar
Programming Basics Web Programming.
Aim #2.2 What else is there to know about Functions and Their Graphs?
The importance of Time Synchronisation in Information Security
Optimal Sorting Algorithms for Parallel Computers
Introduction.
Introduction.
Encodings and reducibility
Fast Refrigerant Property Calculations
CSE8380 Parallel and Distributed Processing Presentation
Week 6 Fatemeh Yazdiananari.
Multi-Pump Control Strategies
Hybrid Programming with OpenMP and MPI
ETL process management with TSQL
Operating System Introduction.
Chemical Kinetics Chemical Reaction Rates (13.2)
Velocity and Other Rates of Change
X y y = x2 - 3x Solutions of y = x2 - 3x y x –1 5 –2 –3 6 y = x2-3x.
Velocity and Other Rates of Change
Potential for parallel computers/parallel programming
Building highly scalable enterprise systems (on Azure)
Chapter 2: Performance CS 447 Jason Bakos Fall 2001 CS 447.
Velocity and Other Rates of Change
EP208 Computational Methods in Physics
Recapitulation of Lecture 12
Introduction to research computing using Condor
Presentation transcript:

Excitable Cell Simulation In Star-P Danilo Šćepanović

Aims Do everything in Star-P so that coding time is low. This is important in experimental work because results often point to changes that have to be made in the code, so coding should be quick Specific Aims Implement an excitable cell model in Matlab Evaluate serial performance on my laptop Evaluate parallel performance for speed and accuracy Summarize recommendations for future May 13, 2009 2

The Model The model was implemented. It is a system of 29 coupled ordinary differential equations. Serial simulations show that it is a stiff system. Ode45 is about 200 times slower than ode15s in Matlab, with negligible difference in the solution May 13, 2009 3

Parallel Simulations I want to run multiple simulations with different initial conditions. The initial conditions are points from the solution over one period This gives time-shifted simulations Allows for high-resolution computation of instantaneous heart rate Stepwise external current was used to drive the heart rate May 13, 2009 4

Speed Parallel Each processor (1, 2, 4, 8 processors) runs 10 simulations Serial The processor runs 10, 20, 40, or 80 simulations Serial time increases linearly with # of simulations Parallel time stays constant! May 13, 2009 5

Accuracy Relative Error: abs(parallel-serial)/mean(abs(serial)): Mean relative error ~2% Minimum relative error 0 Maximum relative error ~100% Why such a large maximum error? May 13, 2009 6

Accuracy At start of simulation, solutions are identical At end of simulation, solutions have drifted! Solutions are identical for first 700ms, then they go out of synch. Must be due to processor differences May 13, 2009 7

Summary Star-P did not have a stiff solver implemented when I began the project, so one was quickly put together for me on the ISC parallel machine Godzuki (thanks to Viral Shah at ISC!) Long simulations, or large numbers of simulations would crash on the parallel computer My laptop ran arbitrarily long simulations without a problem. The code is error-free; problem seems to be with ode solver on Godzuki Coding/debugging in Matlab is easy; debugging errors that occur on the server when running in parallel is hard This problem distributes nicely, N processors cause N-fold speedup Different processors yield slightly different solutions; beware of distributed problems on clusters with dissimilar processors Taking all factors into consideration, the recommendations are: Jobs taking ~8 hours to run should be done in serial Jobs taking days should go to a parallel machine to decrease runtime to ~8 hours. Save results often due to possibility of crashes! May 13, 2009 8