This is a Github repository for the SIGCOMM'23 paper "Network Load Balancing with In-network Reordering Support for RDMA".
This repository incudes p4 programs of ConWeave.
We used BF-SDE-9.11.1
to compile and run the program.
leaf_conweave_resource
is a repo to solely evaluate the data-plane resource consumption of ConWeave mechanism.leaf_conweave_resource/mau.resources.log
is the output log files, showing percentages of resource consumption based onp4-build
.
leaf_conweave
is a repo of ConWeave p4 source code running on leaf (ToR) switches.
NOTE: The current repository is simply provided as a reference code. It would be hard to exactly reproduce the testbed setup and evaluation results in the paper because of our complex testbed environment (e.g., by virtualized topology and its adapted codebase).
For artifact evaluation, feel free to skip this as the majority of results in the paper are executed by RDMA NS-3 simulator that allows integrating various environment conditions that are hard to be done on physical testbed, and prevent randomness for fair comparative studies versus baseline existing solutions.
If time allows, we will provide a simplified / portable program that is runnable on much simpler testbed and provides easy reproducibility.
Some Tofino2 users have difficulty to use the feature of queue pause/resume using AFC. To this end, we provide a toy example so that you can easily run AFC and test whether the queue is indeed paused or not.
For RDMA testbed setup, you need to configure p4 switch with ECN-marking for DCQCN or DCTCP. We provide an example script of DCQCN implementation on Tofino1.
@inproceedings{song2023conweave,
title={Network Load Balancing with In-network Reordering Support for RDMA},
author={Song, Cha Hwan and Khooi, Xin Zhe and Joshi, Raj and Choi, Inho and Li, Jialin and Chan, Mun Choon},
booktitle={Proceedings of SIGCOMM},
year={2023}
}