Kayak

Kayak is a prototype for proactive-adaptive arbitration between shipping compute and shipping data. Kayak is a fork of Splinter.

Setting up Kayak

To run Kayak on CloudLab using xl170 machine, follow these steps:

Instantiate a CloudLab cluster using the python profile cloudlab_profile.py. This should provide a cluster of bare-metal machines running Ubuntu 18.04 with kernel 4.15.0-147 and gcc 7.5.0.
Install MLNX OFED driver. For compatibility, check here. The LTS version is recommended.
Run the provided setup.sh. This will install DPDK as well as the latest version of Rust.
make.
Create splinter/client.toml for client configuration and db/server.toml for server configuration.

You can also refer to the README.md of splinter here or here.

Running Baseline

To run the YCSB-T workload: Inside splinter/, run sudo env RUST_LOG=debug LD_LIBRARY_PATH=../net/target/native ./target/release/ycsb-t

To run the synthetic workload: Inside splinter/, run sudo env RUST_LOG=debug LD_LIBRARY_PATH=../net/target/native ./target/release/pushback

Running Kayak

To run the YCSB-T workload: Inside splinter/, run sudo env RUST_LOG=debug LD_LIBRARY_PATH=../net/target/native ./target/release/ycsb-t-kayak

To run the synthetic workload: Inside splinter/, run sudo env RUST_LOG=debug LD_LIBRARY_PATH=../net/target/native ./target/release/pushback-kayak

Notes

please consider to cite our paper if you use the code or data in your research project.

@inproceedings{kayak-nsdi21,
  author    = {Jie You and Jingfeng Wu and Xin Jin and Mosharaf Chowdhury},
  booktitle = {USENIX NSDI},
  title     = {Ship Compute or Ship Data? Why Not Both?},
  year      = {2021}
}

Acknowledgements

Thanks to Chinmay Kulkarni, Ankit Bhardwaj and Ryan Stutsman for the Splinter repo.

Contact

Jie You (jieyou@umich.edu)