SMP-Parallel-Execution-Framework

Created: 2012-03-22 02:00
Updated: 2019-03-06 16:10

README.md

Simple framework for executing massively parallel tasks. It supports pluggable tasks and pluggable task schedulers. By default, it allows for parallel console commands scheduled in a queue, where the number of tasks running in parallel is equal to the number of available cores.

Compilation

make

SMP Usage

FileAnalysisFramework works like so (executed from the same directory where everything was made):

java parallel.smp.FileAnalysisFramework [-cores #cores to use] "command that takes a file name" files-to-process

...for example:

java parallel.smp.FileAnalysisFramework "./foo" file1.txt file2.txt file3.txt

The above command like will execute:

./foo file1.txt
./foo file2.txt
./foo file3.txt

These will be executed in parallel where possible (due to limitations of the number of cores). Since output isn't captured, all output must be done to files. More specifically, -cores specifies the number of threads to use, which defaults to the number of cores available.

The default queue scheduler doesn't provide any guarantees regarding execution order. While it is true that file N must be selected for processing before file N + 1, file N + 1 may end up being processed first depending on how the JVM schedules threads. If any guarantees are needed regarding execution order, it will be neccessary to implement your own scheduler.

Cluser Usage

There is a master/slave architecture with a very basic protocol for exchanging files. The master serves input files, receives output files, and keeps track of which files have been analyzed. If the master dies, then the analysis comes to a halt. If a client dies, this is not a huge issue; this won't bring the master down and any file that the client was analyzing will be ditributed again later.

The master can be started like so:

java parallel.cluster.FileServer files

...this will set up analysis to work with the given files. The command will block until all files have been analyzed.

Clients can be started like so:

java parallel.cluster.DistributedFileAnalysisFramework master_address command

...where master_address is the IP or DNS-assigned name for the master. The command has additional constraints placed upon it, since output files are sent to the master. The command is given a filename to analyze, and it MUST produce an output file of the form filenamegiven.output in the SAME directory as the input file. The given filename will be a complete path. Note that at the time of this writing, filenames are NOT preserved, so the analysis cannot be sensitive to filenames.

Cookies help us deliver our services. By using our services, you agree to our use of cookies Learn more