/nopaw

Primary LanguagePython

nopaw- just a pipeline

  • TODO write basic test routines: read/write - (live) execution unit test with runme followed by verifyrun verifyrun - use pre-existing read-10/write-10 to ensure verify methods continue to give expected result analyzerun - use pre-existing read-10/write-10 to ensure analysis methods continue to give expected result
  • TODO new operation types to test file writing via: bash wrapper, python script (+subprocess)
  • TODO same order of positional arguments in runme and verifyrun
  • TODO option to switch between gridfs based file data and the in-task-document model
  • TODO "pipeline" executors with read/write operations

Simple pipeline for database read and write tests. A new database is created, and N parallel read/write operations take place, plus a 60 second sleep. The sleep is sort of a stand-in for the (still TODO) pipeline executors.

Simple wrappers are used to place loop-free communicators onto your HPC in an LRMS job. The communicators are told to do a read or write operation when you launched the job, and perform the operation after they are distributed on the HPC, then shutdown following their sleep.

nopaw execution sequence

The reads or writes can then be verified for data integrity. A number of timestamps are reported, which can be analyzed to understand the maximum possible performance of a Workflow Management System built on the particular database in the tested HPC environment.

There are a small number of control parameters that can be varied to understand how the performance changes in response to expectable stressors:

  • Scale: n replicates
  • Operation Type: read or write
  • Data Size: size and unit (blank=k, m)
  • Layout Factor: communicators per node

Currently nopaw MongoDB as database and PyMongo to interface. Could easily provide wrappers for simple operations with other databases, or exchange the PyMongo connector/Python Executor with other interfaces such as C++, Java, etc.

Install:

git clone https://github.com/pawtool/nopaw.git
cd nopaw
./install.sh

Tests:

Run these to check that your platform installed correctly. Note that you must correctly specify a small number of runtime system details in the configuration file nopaw.yml.

tests/read10
tests/write10

Steps:

  1. Copy (and edit if necessary) existing LRMS config file to match your resource
  2. runme to configure and launch job
  3. wait for LRMS job
  4. verifyrun to verify all operations
  5. analyzerun to process timestamp data

Usage:

Always source the RC file: 'rt.bashrc'

Run Script Usage:

runme [runname] [nreplicates] [operation: "read"/"write"]
    optional: [datasize]

Verify Script Usage:

verifyrun [runname] [operation: "read"/"write"] [nreplicates]
    optional: [dbhost] [dbport] [dbname] [datasize]

TODO Analyze Script Usage:

analyzerun [runname]

Example Run-and-Verify:

From the directory you cloned for install, just use these simple steps to run nopaw. These examples are set up identical to the preconfigured tests.

source rt.bashrc
# Test the write operation
runme write10 10 write
# read version
runme read10 10 read

< ... wait for run to complete ... >

verifyrun write10 write 10

Example Run Weak Scaling of Writes:

source rt.bashrc
scales=( 10 40 120 250 500 1000 2008 4016 8048 16000 )

for scale in "${scales[@]}"
do
  runme weak-w-$scale $scale write
done

< ... wait for jobs to complete ... >

for scale in "${scales[@]}"
do
  verifyrun weak-w-$scale write $scale
  analyzerun weak-w-$scale write $scale
done