Dear All,
This draft explores a timely intersection of networking and AI. It identifies significant network challenges in distributed LLM inference clouds—specifically highly concurrent model loading and severe cold start latency caused by massive model downloads (70GB to 1TB+). We analyze why these scenarios are a perfect fit for multicast and discuss the applicability of PIM-SM, SR P2MP, and BIER technologies. We believe this work can serve as a useful discussion starter on how to evolve multicast to better serve emerging AI workloads. We would be very grateful for your review comments and feedback on the draft. Your insights will be invaluable in helping us improve its technical depth and clarity in the next revision. Thank you for your time and consideration. Best regards, Yisong Liu ----邮件原文----发件人：internet-drafts <[email protected]>收件人：Junye Zhang <[email protected]>,Yisong Liu <[email protected]>,Zheng Zhang <[email protected]>抄 送: (无)发送时间：2026-02-12 16:52:29主题：New Version Notification for draft-liu-rtgwg-llmsync-multicast-00.txtA new version of Internet-Draft draft-liu-rtgwg-llmsync-multicast-00.txt hasbeen successfully submitted by Yisong Liu and posted to theIETF repository.Name: draft-liu-rtgwg-llmsync-multicastRevision: 00Title: Multicast Use Cases for Large Language Model SynchronizationDate: 2026-02-12Group: Individual SubmissionPages: 6URL: https://www.ietf.org/archive/id/draft-liu-rtgwg-llmsync-multicast-00.txtStatus: https://datatracker.ietf.org/doc/draft-liu-rtgwg-llmsync-multicast/HTML: https://www.ietf.org/archive/id/draft-liu-rtgwg-llmsync-multicast-00.htmlHTMLized: https://datatracker.ietf.org/doc/html/draft-liu-rtgwg-llmsync-multicastAbstract: Large Language Models (LLMs) deployments are becoming increasingly widespread, with inference services being the most common application. This draft will discuss multicast use cases for inference cloud services.The IETF SecretariatSubject：New Version Notification for draft-liu-rtgwg-llmsync-multicast-00.txtA new version of Internet-Draft draft-liu-rtgwg-llmsync-multicast-00.txt hasbeen successfully submitted by Yisong Liu and posted to theIETF repository.Name: draft-liu-rtgwg-llmsync-multicastRevision: 00Title: Multicast Use Cases for Large Language Model SynchronizationDate: 2026-02-12Group: Individual SubmissionPages: 6URL: https://www.ietf.org/archive/id/draft-liu-rtgwg-llmsync-multicast-00.txtStatus: https://datatracker.ietf.org/doc/draft-liu-rtgwg-llmsync-multicast/HTML: https://www.ietf.org/archive/id/draft-liu-rtgwg-llmsync-multicast-00.htmlHTMLized: https://datatracker.ietf.org/doc/html/draft-liu-rtgwg-llmsync-multicastAbstract: Large Language Models (LLMs) deployments are becoming increasingly widespread, with inference services being the most common application. This draft will discuss multicast use cases for inference cloud services.The IETF Secretariat
_______________________________________________ rtgwg mailing list -- [email protected] To unsubscribe send an email to [email protected]
