Skip to content

A Benchmark Harness for Systematic and Robust Evaluation of Streaming State Stores

License

Notifications You must be signed in to change notification settings

CASP-Systems-BU/Gadget

Repository files navigation

Gadget: A New Benchmark Harness for Systematic and Robust Evaluation of Streaming State Stores

This repository contains instructions for reproducing the experiments in our EuroSys'22 paper "A New Benchmark Harness for Systematic and Robust Evaluation of Streaming State Stores".

Gadget is a benchmark for an accurate evaluation of KV stores for stateful streaming applications. Gadget supports one or more configurable data sources and simulates the internal operations of a stream processing system to generate realistic state access workloads. Gadget generates and issues state access requests to the KV store while collecting performance measurements on latency and throughput. Gadget currently supports four KV stores with different design and performance characteristics: RocksDB, Lethe, FASTER, and BerkeleyDB.

Table of Contents

Project Structure

> tree .
├── configs                 # Instructions on configuring Gadget
├── experiments             # Instructions on  performing experiments with Gadget
├── replayer                # replay ycsb trace on KV stores
├── reproduceEuroSysResults # reproducing EuroSys22 paper results
├── src                     # Gadget Source Code
├── trces                   # Google borg traces
└── tests                   # Unit tests 

Description & Requirements

How to access

Gadget can be downloaded from the following link: https://github.com/CASP-Systems-BU/Gadget

Hardware dependencies

We used the following hardware for our experiments: We run all experiments on a dual-socket machine equipped with 12-core Intel Xeon 4116 CPU running at 2.1 GHz, 32 GB of RAM, and 512GB PC400 NVMe(SK hynix) hard disk.

Software dependencies

We use Ubuntu 20.04 (Linux kernel version5.4). To evaluate a KV store performance using Gadget, you need to install the KV store on your system. To make this step straightforward, we provide a container that already has all KV stores discussed in this paper installed.

Traces

In our experiments, we use traces from Google's Borg system. These traces can be found here

Reproducing the EuroSys22 results

For reviewers, we have prepared a CloudLab profile and a container image (with Gadget and KV stores). Please see a detailed description of reproducing the results presented at the EuroSys22 conference here.

Install and Run Gadget

If you wish to compile Gadget and try it out on your system. You can follow the following steps:

  1. install build essentials (e.g., compiler):

sudo sh install_build_essentals.sh

  1. compile Gadget:
    git clone [email protected]:CASP-Systems-BU/Gadget.git
    cd Gadget
    mkdir build 
    cd build
    cmake ..
    make
    cd src
  1. To run gadget, you can use one of our config files provided in configs folder(here).

    Run Gadget by:

    ./gadget config.txt.

    This example generates the state access workload for a tumbling window operator. The results go to a file named gadget.log. Please note that to evaluate a KV store using Gadget, the KV store must be installed on the system. (please see here for installing KV stores).

Configure Gadget

Please see here for a detailed description of the Gadget configuration.

Experiments with Gadget

Please see here for a detailed description of performing experiments with Gadget.

Gadget Source Code

Please see here for a detailed description of the Gadget source code.

Conference Paper

[EuroSys] A New Benchmark Harness for Systematic and Robust Evaluation of Streaming State Stores. Esmail Asyabi, Yuanli Wang, John Liagouris, Vasiliki Kalavri, Azer Bestavros. Proceedings of the 17th European Conference on Computer Systems, Rennes, France, April, 2022.

License

Gadget is released under Apache License 2.0

About

A Benchmark Harness for Systematic and Robust Evaluation of Streaming State Stores

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •