**Session Date/Time:** 04 Nov 2024 13:00

```markdown
# hpwan

## Summary

This Boff (Birds of a Feather) session explored high-performance wide area networks (HPWANs), focusing on transport-related issues and identifying potential areas for IETF work. The session included presentations on the state of the art in congestion control, use cases from Google (F&Go) and CERN (WLCG), and perspectives from China Mobile. Open mic discussion followed the presentations, with a focus on scope of possible work items.

## Key Discussion Points

*   **Definition of HPWAN:** The working definition of HPWAN, taken from the draft, was reviewed. It emphasizes high speed, low latency, and ultra-high volume applications in research, academia, and large-scale data processing environments.
*   **Limitations of Cubic:** Cubic, the default congestion control algorithm in Linux, was presented as not being suitable for HPWAN environments due to its poor scalability, sensitivity to loss, and inability to scale with Bandwidth Delay Product (BDP).
*   **Alternative Congestion Control:** BBR was suggested as an alternative, but is seen as complex and not easy to fix.
*   **F&Go (Google):** A presentation covered Google's internal system (F&Go) for copying files across its distributed storage infrastructure, emphasizing decentralized control and fairness.
*   **WLCG (CERN):** The CERN experiment use case highlighted the distribution of large datasets over research and education (R&E) networks, the importance of network principles, and the need for preparing for a significant ramp-up in data volumes with the high-luminosity LHC.
*   **China Mobile:** China Mobile shared its experience providing HPWAN services, emphasizing the impact of CPU overhead on end-to-end throughput, and proposing the exploration of RDMA over WAN.
*   **RDMA over WAN:** The potential and challenges of RDMA over WAN were discussed, particularly regarding the need for lightweight, reliable, and congestion-controlled transport mechanisms. The suitability of IWARP vs. Rocky V2.
*   **Technical Issues:** Open technical issues were identified, including per-flow congestion control vs. multi-flow coordination, reactive vs. preemptive intervention, qualitative vs. quantitative feedback, and static vs. dynamic service scheduling.
*   **Long Distance & High Latency:** High end-to-end latency as a problem for WAN performance was raised as an important consideration.

## Decisions and Action Items

*   **Mailing List:** Encourage further discussions on the HPWAN mailing list.
*   **Internet Drafts:** Encourage the submission of Internet Drafts on specific aspects of HPWAN.
*   **Area Director:** Chairs to discuss the meeting outcomes and possible future steps with the Area Director.
*   **Side Meeting:** A side meeting was scheduled for Thursday to further discuss the implications or potential cooperation with the routing layer.
*   **Collaboration:** Researchers were encouraged to measure and analyze the performance of existing and proposed solutions in real-world HPWAN environments. Data sharing at ICCRG.

## Next Steps

*   Further discussion of HPWAN requirements, challenges, and potential solutions on the mailing list.
*   Exploration of potential IETF work items based on the identified gaps.
*   Discussion with Area Director about the future of HPWAN-related work in the IETF.
```