Sparkpp

A na(t)ive proof-of-concept implementation of Apache Spark in C++.

Compiled & tested under gcc-9.2.1, boost-1.71, cmake-3.15.

Inspired by rust Spark implementation native_spark and based on Spark-0.5.

Example

Check examples

Prerequisites

Check bin/prepare.sh

• Boost
• • Serialization
• • Asio
• • Beast
• Cap'n Proto
• fmt
• gperftools
• • tcmalloc
• • google-gprof
• concurrentqueue
• phmap

Installation

# install
./bin/prepare.sh                 # root
./bin/check.sh                   # check installation version
# env
export SPARK_LOCAL_IP=<local ip>
export CPUPROFILE=<profile file> # if google-gprof enabled
export CPUPROFILESIGNAL=<sig>    # if google-gprof enabled
# master
./bin/start_master.sh
# slave
./bin/start_slave.sh

TODOs

• More precise concept control
• Async network support (-fcoroutines, boost::asio::io_service::async_accept), replacing raw socket + thread_pool
• Compare single boost::serialization without Cap'n Proto (& with boost flags, like no_headers)
• Add config (master/slave addr/port) file support
• new version of Spark optimizations: ShuffleWriter
• See other TODOs in files

Random Thoughts

Check miscs/discussion.md and miscs/report.pdf

Reference

• Spark: Cluster Computing with Working Sets
• Resilient Distributed Datasets: A Fault-Tolerant Abstraction for In-Memory Cluster Computing
• Spark-0.5
• native_spark