Parallel MultiQueue

This is a study of trying to improve the concurrency of a concurrent queue structure. The idea revolves around keeping several queue structures in a ring buffer, and trying to give each thread access to exactly one at a time. The hope is that with multiple threads working on different queues, contention between threads is lessened.

Directory Layout

• bin/ Compiled code (linkable objects and executable binaries)

• doc/ Documentation used to research topic

• paper/ Latex source for papers

• src/ Source code

• tests/ Testing scripts and data from various test runs

Build Information

Dependancies

• Only tested on 64 bit Linux, may not work elsewhere
• Relies on librealtime (Linux specific, and maybe BSD) to accuately time tests
• Only compiles with GCC as it uses GCC's wrappers to x86 atomic primatives (compiled with version 4.6)
• Needs libboost-thread

Building and Running

Just run make to compile the source. This should generate some executable binaries in the bin/ directory which correspond to testing binaries for a two-lock queue and a multiqueue. Each executable runs a test which measures the throughput of a queue under load from a given number of producer and consumer threads. They accept the following flags:

-l       : use large item test (default false)
-s       : use small item test (default true)
-p <num> : specify number of producers (default 2)
-c <num> : specify number of consumers (default 2)
-t <num> : specify time to run in seconds (default 5)
-h       : prints the help text
-v       : verbose

The default behavior is to run the test and print to stdout the throughput of the queue in items per second.

Testing Scripts

The tests directory contains two Python scripts which automatically tests an executable under various thread configurations.

logger.py takes three arguments:

1. Name of an executable from the bin/ directory
2. Number of consumer threads to test up to
3. Number of producer threads to test up to

It will generate a data file in the tests/data/ directory. The file is named for the datetime the script was run at.

scale.py takes a variable number of data files produced by logger.py and produces an equal number of gnuplot scripts which, when run, generate an SVG file which graph the throughput of each passed file. The generated graphs are scaled to the maximum of all logged values in the passed file arguments. The scripts are generated in the tests/scaled/ directory.

License

Copyright (c) 2013 Carlos Valera

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.