KVPaxos is a key-value distributed storage system that uses Paxos and Parallel State Machine Replication to ensure consistency among replicas. It's developed as a prototype to measure latency and throughput when using state partitioning and balanced graph partitioning to schedule requests among threads, it includes 4 graph repartition algorithms to be used during execution: METIS, KaHIP, FENNEL and ReFENNEL.
KVPaxos is a prototype and so it does not cover many corner and common cases, it should not be used as it is in a real deploy context, but can be used as a starting point to other projects.
The experiments where executed in Ubuntu 22.04 64 bits. To make sure you have all dependencies needed, run:
apt install make cmake libtbb-dev libmsgpack-dev libevent-dev mpich pip libboost-all-dev git -y
pip install conan==1.59
Also, install LibCDS 2.3.3 from here, and make sure Conan executable path is present in PATH environment variable.
To compile, run compile_all.sh script. After that, scripts for each experiment can be executed in build/bin foulder
CMake is used to build the project, along with Conan to control dependecies. Conan downloads the following dependencies:
- LibEvent 2.1.11
- toml11
- tbb 2020.1
Dependencies not present in Conan are added as submodules, so make sure to recursively clone submodules too when cloning the project. For some reason that I've not cracked, libpaxos in deps/libpaxos is often cloned in an older commit, a git checkout master in the directory may be necessary.
Conan only controls dependecies that are directly used by KVPaxos, dependecies used by submodules need to be installed separately. The submodules and their dependecies are:
In build/bin execute one of the scripts. The script generates the workload and executes the tests of a given implementation under a given workload.
The build/bin/output directory will contain the results of the experiments, divided in subfolders. Each experiment creates two CSV files, one containing throughput information along the execution, and other, prefixed with "details_", containg information about the scheduler and partitioner.