A Queueing Network Model for Performance Prediction of Apache Cassandra

<p>The dataset consists in several csv files containing Cassandra and ScyllaDB performance.</p> <p>The experiments are organized in folders. There are three main folders containing:<br> - Cassandra 4 nodes: the files related to the Cassandra experiments conducted on a cluster composed of four nodes.<br> - ScyllaDB 4 nodes: The files related to the ScyllaDB experiments conducted on a cluster composed of four nodes. <br> - Cassandra QUORUM variant: the simulation data where a different kind of QUORUM is implemented in Cassandra.<br> <br> "Cassandra 4 nodes" and "ScyllaDB 4 nodes" include some subfolders, each one containing the files of the Consistency Level applied for those experiments. Each experiment is composed by three files (data*.csv) with the data reported by Yahoo! Cloud System Benchmark (YCSB) in the end of the experiment execution. Each folder contains also a sim.csv file with the data gathered from the simulation of the model inside Java Modeling Tool.</p> <p>The data*.csv files are composed by:<br> -Number of threads or clients<br> -Overall Throughput<br> -Number of Read requests<br> -Overall Read Response Time<br> -95 percentile Read Response Time<br> -99 percentile Read Response Time<br> -99.9 percentile Read Response Time<br> <br> Differently, the sim.csv files are composed by:<br> -Number of threads or clients<br> -Overall Throughput<br> -Overall Read Response Time