In-Network Collective Operations: Game Changer or Challenge for AI Workloads?
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911401852796928 |
|---|---|
| author | Hoefler, Torsten Khalilov, Mikhail Clark, Josiah Anubolu, Surendra Kalkunte, Mohan Schramm, Karen Spada, Eric Roweth, Duncan Underwood, Keith Caulfield, Adrian Kabbani, Abdul Rastegari, Amirreza |
| author_facet | Hoefler, Torsten Khalilov, Mikhail Clark, Josiah Anubolu, Surendra Kalkunte, Mohan Schramm, Karen Spada, Eric Roweth, Duncan Underwood, Keith Caulfield, Adrian Kabbani, Abdul Rastegari, Amirreza |
| contents | This paper summarizes the opportunities of in-network collective operations (INC) for accelerated collective operations in AI workloads. We provide sufficient detail to make this important field accessible to non-experts in AI or networking, fostering a connection between these communities. Consider two types of INC: Edge-INC, where the system is implemented at the node level, and Core-INC, where the system is embedded within network switches. We outline the potential performance benefits as well as six key obstacles in the context of both Edge-INC and Core-INC that may hinder their adoption. Finally, we present a set of predictions for the future development and application of INC. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_19132 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | In-Network Collective Operations: Game Changer or Challenge for AI Workloads? Hoefler, Torsten Khalilov, Mikhail Clark, Josiah Anubolu, Surendra Kalkunte, Mohan Schramm, Karen Spada, Eric Roweth, Duncan Underwood, Keith Caulfield, Adrian Kabbani, Abdul Rastegari, Amirreza Networking and Internet Architecture Artificial Intelligence Hardware Architecture Performance Systems and Control This paper summarizes the opportunities of in-network collective operations (INC) for accelerated collective operations in AI workloads. We provide sufficient detail to make this important field accessible to non-experts in AI or networking, fostering a connection between these communities. Consider two types of INC: Edge-INC, where the system is implemented at the node level, and Core-INC, where the system is embedded within network switches. We outline the potential performance benefits as well as six key obstacles in the context of both Edge-INC and Core-INC that may hinder their adoption. Finally, we present a set of predictions for the future development and application of INC. |
| title | In-Network Collective Operations: Game Changer or Challenge for AI Workloads? |
| topic | Networking and Internet Architecture Artificial Intelligence Hardware Architecture Performance Systems and Control |
| url | https://arxiv.org/abs/2601.19132 |