This project contains data, tools, and scripts required to execute workloads and analyses intended for evaluating the Self-aware Resource Demand Estimation (SARDE) project. SARDE enables continuous resource demand estimations, while constantly adapting and optimizing its used estimators. SARDE is designed as an AddOn to the classic LibReDE library.
Check out the source code of SARDE here, and LibReDE here for more information.
The main publication for SARDE is currently under review at ACM Transactions on Autonomous and Adaptive Systems (TaaS).
This project currently contains four folders:
-
librede-artificial_exp: Contains the micro-benchmarks data set. This data set consists of a set of measurements obtained by executing micro-benchmarks on a real system. A set of 210 traces, each with approximately one hour run-time, was collected. The micro-benchmarks generate a closed workload with exponentially distributed think times and resource demands. The think times themselves were set to fit the targeted load level of each specific experiment. As mean values for the resource demands, we selected 14 different subsets of the base set (0.02s, 0.25s, 0.5s, 0.125s, and 0.13s) with a varying number of workload classes (1, 2, and 3) and target load levels (20%, 50%, and 80%) arbitrarily chosen from the base set. This way, we can ensure that the resource demands are not linearly growing across workload classes. Additionally, the subsets intentionally contained cases where two or three workload classes had the same mean resource demand. The folder contains only experiments sampled every 1 second, then splits into the three different loads, each with 14 different experiment setting, which were each repeated 5 times. In addition, a script for removing unnecessary files is provided, that might get created while experimenting on the data.
-
librede-parsing: Contains a parse script that parses log-data produced by the monitoring and the workload generator into data that can be analyzed and processed by LibReDE. In addition, it contains the analysis data for the recent experiment run.
-
librede-synthetic: Contains the workload driver, the load script, as well as the intensity description for measuring and setting the synthetic application under load. In addition, the output of a set of previous experiment runs is stored. A README provides instructions on how to execute the respective benchmark.
-
librede-analysis: Contains analysis scripts for analysing the performance of SARDE using the so-called
logbooks, i.e., the output of a continuous experiment run conducted by SARDE. In addition, some logbooks from previous experiments are contained.
The majority of the analysis and the respective data from the main experiments of the paper are also published as a CodeOcean Capsule for an easy one-click reproducibility, without the need for a setup.