Authors: John Albury, Dax Borde, Matthew Garrison, Richard Snyder
In this project, we re-implement the lock-free vector described in Scalable Lock-Free Vector with Combining. Additionally, we implement a transactional version of this vector using the Rochester Software Transaction Memory (RSTM) library. Finally, we compare the performance of our two implementations to each other.
You can view the final report here.
First, download the main.cpp file from Webcourses and download RSTM v7 from https://code.google.com/archive/p/rstm/downloads.
Next, unpack rstm to somewhere like /home/rstm and run the following from /home or wherever you unpacked rstm.
mkdir rstm_build
cd rstm_build
cmake ../rstm
make
This should build RSTM on your system. If it doesn't work, try running it on an Ubuntu 18.04 VM; that's what I ended up using.
Next, test that everything is working by copying main.cpp to rstm/bench. In rstm/bench/CMakeLists.txt, add main to the list starting on line 22. Then, run:
cd rstm_build
make
bench/mainSSB64
It should output something like x = 40000.
Now, move the stm_vector.cpp and stm_vector.h files to rstm/bench (you can remove main.cpp from there now). In rstm/bench/CMakeLists.txt, replace main with stm_vector in the list starting on line 22. Also, move run_stm.py to rstm_build/. The following is how to run version 2 (our modified version of the STM vector). To run the unmodified version, simply change bench/stm_vector_v2SSB64 to bench/stm_vector_v1SSB64
cd rstm_build
make
python3 run_stm.py bench/stm_vector_v2SSB64
The results will be located in the directory you end up in (rstm_build/) under the name results.txt. If you re-run the test, make sure to delete or move results.txt before re-running, as the output will be appended to results.txt if it already exists.