In-network Aggregation Research

Representative Groups

• Wenfei Wu
• Lin Wang

Open Sourced Codes

• SwitchML
• Open source P4 implementations

Paper List

Published in 2023

In this year, some works are proposed to simplify the development and expand the application scenario of in-network aggregation.

• ClickINC: In-network Computing as a Service in Heterogeneous Programmable Data-center Networks, SIGCOMM 23, Wenfei Wu
• NetRPC: Enabling In-Network Computation in Remote Procedure Calls, NSDI 23, Wenfei Wu
• A Generic Service to Provide In-Network Aggregation for Key-Value Streams, ASPLOS 23 (Distinguished Paper Award), Wenfei Wu
• In-Network Aggregation with Transport Transparency for Distributed Training, ASPLOS 23, Wenfei Wu

Additionaly, some works are proposed to improve the performance of in-network aggregation.

• A2TP: Aggregator-aware In-network Aggregation for Multi-tenant Learning, Eurosys 23, Zhaoyi Li at CSU

Published in 2022

In fact, papers listed below do not strictly relate to in-network aggregation. They either discuss the performance of novel programmable chip under INA workloads (Trio) or propose works on running tasks related to DNN with programmable switches.

• Using trio: juniper networks’ programmable chipset - for emerging in-network applications, SIGCOMM 22, Mingran Yang at MIT
• Taurus: a data plane architecture for per-packet ML: ASPLOS 22, Tushar Swamy at Stanford: Proposing a FPGA-based prototype to run per packet inference at programmable switches
• Holistic Resource Scheduling for Data Center In-Network Computing,TON , Lin Wang group
• Distributed DNN Serving in the Network Data Plane, EuroP4 22, Lin Wang group

Published in 2021

• ATP: In-network Aggregation for Multi-tenant Learning, NSDI 21 (best paper award), Wenfei Wu group
• Scaling Distributed Machine Learning with In-Network Aggregation, NSDI 21, Microsoft: This work was proposed in 2019, focusing on in-network aggregation with one switch
• In-Network Aggregation for Privacy-Preserving Federated Learning, ICT-DM 2021: Performing federated learning with in-network aggregation
• In-network Aggregation for Shared Machine Learning Clusters, MLSys 2021, Manya Ghobadi at MIT: Focusing on routing protocol with in-network aggregation
• P4COM: In-Network Computation with Programmable Switches, Arxiv, Ge Chen at Alibaba

Some papers focus on improving peformance of in-network computing from resource scheduling or programming language.

• Switches for HIRE: Resource Scheduling for Data Center In-Network Computing, ASPLOS 21, Lin Wang group
• Don’t You Worry 'Bout a Packet: Unified Programming for In-Network Computing, HotNets 21, Lin Wang group

Published in 2019

• Accelerating Distributed Reinforcement Learning with In-Switch Computing: ISCA 19, Youjie Li at UIUC

Eariler

• DAIET: a system for data aggregation inside the network, SoCC 17: Preliminary work of SwitchML
• Scalable Hierarchical Aggregation Protocol (SHArP): A Hardware Architecture for Efficient Data Reduction,COMHPC 16, NVIDIA: Proposing a dedicated switch architecture to perform in-network aggregation (reduction)
• NetAgg: Using Middleboxes for Application-specific On-path Aggregation in Data Centres, CoNEXT 14: Implement in-network aggregation via switch-attached high-performance middleboxes
• Camdoop: Exploiting In-network Aggregation for Big Data Applications, NSDI 12: They use servers in a direct-connect network topology to perform in-network aggregation

Summary — Jul 3, 2023