# More than bin packing: Dynamic resource allocation strategies in cloud data centers

# References

- monitoring.zip contains an CSV export of the monitoring data gathered in our infrastructure for the experiments reported in this paper. See section Experiments. 

- hdfs.tar.gz contains an HDFS dump of our whole Sonar database. This data was actually used by all our analysis scripts to generate the results found in this paper. It not just covers monitoring data over the last year that is relevant to this work but also monitoring data that was used in multiple earlier experiments that were published in different works. 

- times.tar.gz contains the time series data used by simulations and experiments. Time series are encoded using Apache Thrift and can be accessed using the Times service (see https://github.com/jacksonicson/times)

## VM schedule (folder schedules)

The folder contains configuration data (configuration.csv) as well as simulation results (sensitivity.csv, sensitivity_lowerbound.csv) that were used to determine the significant parameters of our VM arrival and departure schedules. The schedules/*.json files describe the schedules used by the simulations and experiments. 

## Simulations (folder simulations)

The file simulations.csv contains all data generated by controller simulations in a single CSV file. 

## Experiments (folder experiments, monitoring.zip, hdfs.tar.gz)

The ZIP file monitoring.zip contains RAW monitoring data of all experiments reported in this paper in the form of multiple CSV files. Originally the data was collected using the Sonar monitoring system (see https://github.com/jacksonicson/sonar) which was used to monitor our whole testbed infrastructur. These Sonar source codes should serve as a documentation and describe which monitoring sensors exist and how they work. 

Each CSV files holds all recorded data starting from Unix timestamp 1383264000 until 1425816993. There is one file for each combination of hostname (e.g. server or VM) and sensor (e.g. CPU, memory). In addition there is a set of CSV files with log data for each combination of hostname and log-channel.

The file experiments.csv describes the start and end time for each experiment conducted together with its parameterization and the aggregated data produced by our analysis scripts. Each experiment is fully described by combination of the the Times database (times.tar.gz), the schedules, and the Sonar monitoring data (hdfs.tar.gz, or monitoring.zip).

# R analysis scripts

The R scripts were used to create summary statistics based on the aggregated analysis data of the CSV files.
