Robust Workflows for Science and Engineering

Transcription

1 Robust Workflows for Science and Engineering David Abramson, Blair Bethwaite, Colin Enticott, Slavisa Garic, Tom Peachey Anushka Michailova, Saleh Amirriazi, Ramya Chitters Faculty of Information Technology, Monash University, Clayton, 3800, Victoria, Australia Department of Bioengineering, University of California San Diego, and San Diego Supercomputer Centre 9500 Gilman Drive, La Jolla, USA 1

2 Introduction Robust Science with Nimrod Nimrod/Kworkflows The Tagged Dataflow Architecture Nimrod/K Director New Kepler Actors Some examples from cardiac science

3 Robust science with Nimrod

4 Nimrod supporting real science A full parameter sweep is the cross product of all the parameters (Nimrod/G) An optimization run minimizes some output metric and returns parameter combinations that do this (Nimrod/O) Design of experiments limits number of combinations (Nimrod/E) Workflows (Nimrod/K) Nimrod/O Results

5 Plan File Nimrod Portal Nimrod/O Nimrod/E parameter pressure float range from 5000 to 6000 points 4 parameter concent float range from to points 2 parameter material text select anyof Fe Al task main copy compmodel node:compmodel copy inputfile.skel node:inputfile.skel node:substitute inputfile.skel inputfile node:execute./compmodel < inputfile > results copy node:results results.$jobname endtask Nimrod/G Actuators Grid Middleware 5

6 Sent to available machines Prepare Jobs using Portal Results displayed & interpreted 6 Jobs Scheduled Executed Dynamically

7

8

9 Drug Docking Antenna Design Aerofoil Design 9

10

11 Elastic Resources Resource #jobs completed Total job time (h:m:s) East :37:23 91/5.7 EC :34:05 67/14.2 µ / σ Job runtime (mins)

12 Nimrod/K Workflows

13 Nimrod/K Workflows Nimrod/K integrates Kepler with Massivly parallel execution mechanism Special purpose function of Nimrod/G/O/E General purpose workflows from Kepler Flexible IO model: Streams to files Authentication GUI Kepler GUI Extensions Vergil Documentation Kepler Object Manager SMS Actor&Data SEARCH Type System Ext Smart Re-run / Failure Recovery Provenance Framework Kepler Core Extensions Ptolemy

14 Kepler Directors Orchestrate Workflow Synchronous & Dynamic Data Flow Consumer actors not started until producer completes Process Networks All actors execute concurrently IO modes produce different performance results Existing directors don t support multiple instances of actors. 14

15 Workflow Threading Nimrod parameter combinations can be viewed as threads Multi-threaded workflows allow independent sequences in a workflow to run concurrently This might be the whole workflow, or part of the workflow Tokens in different threads do not interact with each other in the workflow

16 The Nimrod/K director Implements the Tagged Data Architecture Provides threading Maintains copies (clones) of actors Maintains token tags Schedules actor s events Nimrod Director

17 Dataflow Execution 101

18 The MIT/Manchester model

19 Dynamic Parallelism Token Colouring Actor Clone 1 Clone 2 Clone 3

20 Nimrod Director Simple Example

21 Nimrod Director At run time

22 Director controls parallelism Uses Nimrod to perform the execution

23 New Actors to support robust science

24 Complete Parameter Sweep Using a MATLAB actor provided by Kepler Local spawn Multiple thread ran concurrently on a computer with 8 cores (2 x quads) Workflow execution was just under 8 times faster Remote Spawn 100 s 1000 s of remote processes

25 Parameter Sweep Actor

26 Partial Parameter Sweep

27 Nimrod/EK Actors Actors for generating and analyzing designs Leverage concurrent infrastructure

28 Nimrod/E Actors No actor parameters need setting No difference from the parameter sweep actors

29 Parameter Optimization: Inverse Problems Domain Definer Points Generator F(x,y,z,w, ) Optimizer Constraint Enforcer Execute Model

30 Nimrod/OK Workflows Nimrod/K supports parallel execution General template for search Built from key components Can mix and match optimization algorithms 30

31 Cardiac Modelling Experiments Heart disease still leading cause of death Understanding the underlying physiological mechanisms is cheaper and faster when experimental studies are performed together with mathematical models & computer simulations Model parameters often need to be tuned Normal Rhythm Ventricular Fibrillation

32 Experiment 1 Solutions of ODE s Ion movement in single cells After adding the new channel and metabolic factors, confirm the results against physiological data. Stabilize the model for a heart rate of 0.5Hz and a longer performance duration (3 min). In both, the focus is on calcium transient concentrations, and action potential shape and duration, to determine the model stability and accuracy.

33 Experiment 1 In order to validate the model, outputs are compared against experimentally known data. Explore nine input metabolic constants and selecting the combination that produces the closest output to the desired value. This experiment performed simplex searches over the nine-dimensional space.

34 Experiment 1 Results Machine # Cores Hardware Location East 160 Intel(R) Xeon(R) 1.60GHz Monash University, Clayton 64 Intel(R) Xeon(R) 3.00GHz West 160 Intel(R) Xeon(R) 1.60GHz Deakin University, Geelong South 160 Intel(R) Xeon(R) 1.60GHz RMIT, Melbourne Index Objective Jobs Batches Min Average Worst

35 Experiment 2 Examine the changes in which total ionic flux(s) play the largest role in determining specific model tissue layer in rabbits. Use Nimrod/EK to perform a parameter sweep and measure the effects of each of the parameter combinations

36 RESULTS AP_len 1_ADP_90_10percent - Lenth plot H F E F E

37 RESULTS CON T AP_len 1_ADP_90_1 0percent - Daniel plot Ca_cyt_len1_ ADP_90_10p ercent Daniel Plot E H E F F

38 Faculty Members Jeff Tan Maria Indrawan Research Fellows Blair Bethwaite Slavisa Garic Donny Kurniawan Tom Peachy Admin Rob Gray Current PhD Students Shahaan Ayyub Philip Chan Colin Enticott ABM Russell Steve Quinette Ngoc Dinh (Minh) Completed PhD Students Greg Watson Rajkumar Buyya Andrew Lewis Nam Tran Wojtek Goscinski Aaron Searle Tim Ho Donny Kurniawan Funding & Support Axceleon Australian Partnership for Advanced Computing (APAC) Australian Research Council Cray Inc CRC for Enterprise Distributed Systems (DSTC) GrangeNet (DCITA) Hewlett Packard IBM Microsoft Sun Microsystems US Department of Energy 38

39 Questions? More information: