Fling Study

Overview

The Fling study is a set of broadband platform simulations defined by PG&E and PEER researchers and run by SCEC Broadband platform group.

Completed Scenarios

1. Scenario141:
   1. Codebase: URS/URS/URS/URS
   2. Completed: 15 May 2012
   3. Operator: PJM
   4. Results Staged:/home/broadband-01/maechlin/flingstudy/Scenario141
   5. Total Files: 36Mb
   6. outdata: 3.1Gb
   7. Starttime: Mon May 14 22:05:02 PDT 2012
   8. Endtime: May 15 11:20:47 PDT 2012
   9. Cores: 16

Terminology

To help the engineers, seismologists, and computer scientists talk about this computational study, we have defined the following terms for use in broadband calculations:

1. Study - A well-defined set of calculations to be performed. To qualify as a "study", the problem must defined to the level of detail that a knowledgeable user could calculate the total number and type of output files that will be calculated. We are currently conducting the Fling study.

1. Scenario - A scenario refers to a earthquake defined by dip, rake, and depth and a set stations that record that earthquake. A scenario does not define rupture time history, and does not define hypocenter location. The Fling study has 42 scenarios, numbered 101 to 142.

1. Scenario Variation - Earthquake definitions given as a scenario, are elaborated into scenario variations. Scenario variations may differ by initial stress distribution and also by hypocenter location, and possibly by rupture velocity. In the fling study, each Scenario has 30 scenario variations.

1. Simulations - In broadband, a simulation is defined by an invocation of the main program, called run_bbp_2G.py. A simulation is defined by a source description, a station list, and a set of Green's functions to use. In the fling study, we run a simulation for each scenario variation.

1. Seismograms - The smallest unit of work in broadband is the seismogram. In broadband, a seismogram refers to a 3 component seismogram associated with a site. At a minimum, the relation between the source location and the site is known. In many cases, the geographical location of the site is known. In the fling study, the number of seismograms is determined by the number of stations in the station list each simulation. For smaller scenarios, the station lists have 96 stations, for larger they have 138 stations. We will calculate the total number of seismograms and use that as a computation metric for the fling study.

Fling Study Scenario List

scenario_id	magnitude	dip	rake	Ztor	fault_length	fault_width		Priority
								
101	6	90	180	0	14	8		1
102	6.5	90	180	0	24	13		
103	7	90	180	0	68	15		
104	7.5	90	180	0	210	15		
105	7.8	90	180	0	420	15		
106	8.2	90	180	0	470	15		
								
107	6	70	180	0	14	8		2
108	6.5	70	180	0	24	13		
109	7	70	180	0	68	15		
110	7.5	70	180	0	210	15		
111	7.8	70	180	0	420	15		
112	8.2	70	180	0	470	15		
								
113	6	70	180	0	14	8		3
114	6.5	70	180	0	24	13		
115	7	70	180	0	40	25		
116	7.5	70	180	0	100	32		
117	7.8	70	180	0	160	40		
118	8.2	70	180	0	400	40		
								
119	6	90	180	0	14	8		4
120	6.5	90	180	0	24	13		
121	7	90	180	0	40	25		
122	7.5	90	180	0	100	32		
123	7.8	90	180	0	160	40		
124	8.2	90	180	0	400	40		
								
125	6	45	90	0	10	10		5
126	6	45	90	5	10	10		
127	6.5	45	90	0	18	18		
128	6.5	45	90	5	18	18		
129	7	45	90	0	44	23		
130	7.5	45	90	0	126	25		
131	7.8	45	90	0	180	25		
								
132	6	60	90	0	10	10		
133	6	60	90	5	10	10		
134	6.5	60	90	0	18	18		
135	6.5	60	90	5	18	18		
136	7	60	90	0	50	20		
137	7.5	60	90	0	150	20		
138	7.8	60	90	0	200	20

139	7.0	45	90	0	18	18
140	6.5	45	90	0	18	18
141	6.5	45	90	5	18	18
142	7.0	45	90	0	44	23
		

Scenarios 140, 141 and 142 were identified as high priority simulations at a NGA-West2 meeting last Friday.

Running on USC HPCC

A sample set of simulations from the Fling study were run on USC HPCC. The original fling generation scripts, source descriptions, station lists, and batch scripts were copied over from broadband.usc.edu to /home/rcf-104. Then small modifications were made to update paths and block the actual execution of the platform (the platform will be run in a PBS job):

Sample scripts can be found at the following locations. However then are not necessarily used in the following order.

Script	Location	Description	Modified
build_xml.py	/auto/rcf-104/patrices/bbp/batch_tools	Builds XML workflows for a simulation	No
batch_run_bbp.py	/auto/rcf-104/patrices/bbp/batch_tools	Executes BBP workflow	Modified to only write BBP command-lines for simulations to a log for later execution by run_parallel.py. BBP invocations are saved in batch_run_bbp_sims.log and bbp output directory moves are saved in batch_run_bbp_moves.log
run_parallel.py	/auto/rcf-104/patrices/bbp/batch_tools	Helper script to run N programs on a set of M cores	New script
gen_source_input.csh	/auto/rcf-104/patrices/bbp/fling	Generate full study inputs	No
run_bbp-parallel.csh	/auto/rcf-104/patrices/bbp/fling	Originally intended to execute the study with the platform. After modifications, only generates XML and execution lists for run_parallel.py.	Some paths changed, also added ${ROOT_PATH} to some relative path locations to make them absolute paths

General steps for running the Fling study:

• Generate inputs

$ ./gen_source_input.csh

• Generate XML workflows

$ ./run_bbp-parallel.csh

• Create PBS job submission script (example below)
• Submit PBS job to USC HPCC

Example PBS script running the sample Fling simulations on 16 cores:

#!/bin/bash

#PBS -q nbns
#PBS -l arch=x86_64,pmem=2000mb,pvmem=3000mb,walltime=6:00:00,nodes=4:ppn=4
#PBS -V
#PBS -e /home/rcf-104/patrices/bbp/fling/Xml1/Set1/run_set1.err
#PBS -o /home/rcf-104/patrices/bbp/fling/Xml1/Set1/run_set1.out

PYTHONPATH=/home/rcf-104/patrices/bbp/11.2.2/bbp_2g/comps

HOME=/home/rcf-104/patrices/bbp/fling

echo "Jobs start"
date

cd $HOME

python $HOME/Xml1/Set1/run_parallel.py /home/rcf-104/patrices/bbp/11.2.2/setup_bbp_env.sh $HOME
/Xml1/Set1/batch_run_bbp_sims.log $PBS_NODEFILE 1

python $HOME/Xml1/Set1/run_parallel.py /home/rcf-104/patrices/bbp/11.2.2/setup_bbp_env.sh $HOME
/Xml1/Set1/batch_run_bbp_moves.log $PBS_NODEFILE 1

echo "Jobs end"
date

Comparison of Seismogram from Server and Cluster

Simulation	broadband.usc.edu	USC HPCC cluster
10010116
10010129

Certification of USC HPCC Cluster for Broadband Calculations

The verification and validation of the currently released Broadband platform is based on results generated on a SCEC server called broadband.usc.edu. When we move the Broadband platform software, re-build it, and re-run it in a different computing environment, the results the platform produces can be slightly different than results produced on the SCEC server. Differences can come from computing hardware, from operating system characteristics, from compiler version, and other sources.

Before accepting results generated in a new computing environment, we must first certify that the new computing environment produces results that are equivalent to the results from the original server where the platform was originally developed and tested.

To speed up execution of the Fling study, we plan to run it on the USC HPCC cluster, so we must certify that USC HPCC cluster results are valid and comparable to those generated on broadband.usc.edu.

Below are initial results from our initial certification tests. A researcher ran a small subset of the Fling study on the SCEC broadband server. Then, we ran the same subset on the USC HPCC cluster. Below we compare the output seismograms from both runs, showing that the two results are very similar.

In our discussions, we decided that the certification criteria for this study will include a number of small magnitude ruptures, and a number of large magnitude ruptures, which we will post when they are available.