Replication Package

This site provides the replication package for the BTW 2021 paper Silentium! Run–Analyse–Eradicate the Noise out of the DB/OS Stack

NOTE: An archival version of the pre-built docker image, together with a copy of the git repository and the measured data, are available at the DOI 10.5281/zenodo.4602296.

Building the Docker image

• Clone the repository

git clone https://github.com/lfd/btw2021.git

• Build the Docker image from scratch

cd btw2021

docker build -t btw2021 .

• Copy measurement package (for x86_64 Linux) to host (i.e., machine used to build binaries)

docker run --rm --entrypoint cat btw2021:latest /home/build/measure.tar.bz2 > /path/to/measure.tar.bz2

Performing measurements on Linux targets

• Copy measure.tar.bz2 from host to target (i.e., machine used to perform measurements), and untar the content.

  The measurement tarball includes pre-generated TPC-H data files of a few 100KiB. However, realistic measurements are usually performed with GiB of data. This necessitates running the generator on the target system. Create sub-folders tpch_size (for instance, tpch_2.0 for 2 GiB), copy dbgen and dists.dss from the data/tpch subfolder into the directory, and run ./dbgen -s 2.0 to create, for instance, a data set of 2.0 GiB.

• Additionally, two measurement modes are provided: Pre-calibrated sampling that only records values that exceed an upper or lower bound, or a complete measurement the records every latency value (default). Set CALIBRATED to 0 or 1 to perform a (non-) calibrated run.

• Adapt doall.sh (and probably also dispatch.sh) for the desired scenario. Variable DATASET can be set to either tpch or finance. CALIB_T specifies the time used for the calibration run (on data of size CALIB_S, whereas RUN_T and RUN_S specify runtime and data size for the measurement proper.

• Execute caps.sh to set capability cap_ipc_lock (which allows to mlock memory allocations without having to run with root privileges) for the measurement binaries.

• Run doall.sh (after performing any specific configuration settings on your system)

• Visualise using the scenario specific scripts in plot/

Performing measurements on RTEMS on x86_64 targets

• Use the binaries provided in /home/build/dbtoaster/rtems in the Docker image. These are compiled for bare-metal x86_64, and can be booted using grub.

Performing measurements on RTEMS on BeagleBone Black

• Binaries for the BeagleBone black are built by default in the Docker container. Copy the binaries from /home/build/dbtoaster/bbb to an SD-Card, and boot the bbb into U-Boot. Assume a working Linux installation is provided on the internal flash of the bbb that provides an appropriate device tree, enter

   load mmc 0:1 0x80800000 axfinder.img
   load mmc 1:1 0x88000000
   /boot/dtbs/4.1.15-ti-rt-r43/am335x-boneblack.dtb bootm 0x80800000 - 0x88000000`

to run, for instance, the axfinder query from the finance dataset.

Inspecting changes to DBToaster

• See https://github.com/lfd/dbtoaster-backend/tree/btw2021

Pre-Built image and measured datasets

For convenience, the DOI archived artefacts are also available at non-archival locations:

• A pre-built Docker image (including a fixed copy of all required sources) is available at https://cdn.lfdr.de/btw2021/btw2021-docker.tar.bz2 (1.5 GiB compressed/5.7 GiB uncompressed)
• Measured observations for all experiments discussed in the paper (raw data) are available at https://cdn.lfdr.de/btw2021/btw2021-data.tar.bz2 (4.7 GiB compressed/9.1 GiB uncompressed)

Remarks

Please see the paper for a discussion of the structure of our replication package.