/micro-benchmarks

Fork of minha-benchmarks, a suit of tests for the MINHA framework

Primary LanguageJava

micro-benchmarks

This repository contains micro-benchmarks with distributed data races:

  • package microbenchmarks includes a number of micro-benchmarks simulating real-world concurrency bugs of popular distributed applications. The name of each package indicates the application name (m - MapReduce, h - Hadoop) and the issue number in the corresponding bug repository.

  • package dataflaskslite contains an implementation of a peer sampling service named Cylon, used by the gossip-based key-value store DataFlasks. This benchmark can be run under different configurations by setting the values in class dataflaskslite.common.Parameters. In particular, one can tune the number of nodes in the system (NUMPEERS), the number of cycles during which the nodes exchange messages (NUMCYCLES), the shuffling period (INTERVAL), the size of each node's view (VIEWSIZE), and the number of peers exchange in a shuffling (GOSSIPSIZE).

Each benchmark includes two main classes: a standalone version and a Minha version.

Usage

1. Compile:

$ mvn package

2. Run jar:

$ java -jar target/minha-benchmarks-1.0-SNAPSHOT-jar-with-dependencies.jar

At the moment, the jar will execute main class dataflaskslite.test.DFMainMinha. Change pom.xmlto compile the jar file for other main classes.

3. Runtime Trace: Assuming that the Minha framework is correctly configured with the shared variables and message handlers of the target program, after the execution of the jar, a trace file minhaTRACER.log will be placed under folder logs. The trace file contains runtime events (in JSON format) regarding: socket sends/receives, read/write accesses to variables of interest, thread synchronization points.