Reshma Das
Reshma Das is a Principal Engineer at HPE specializing in routing protocols, with a primary focus on BGP design and implementation. With over 15 years of experience in the networking industry, her career has spanned OSPF development, routing infrastructure engineering at Juniper Networks, BGP-EVPN at Brocade, and BGP development at both Juniper and HPE.
She is an editor of RFC 10005, BGP Link Bandwidth Extended Community (IETF Internet Standard, June 2026), which is the subject of this presentation, and a co-author of its companion use-cases document, draft-ietf-bess-ebgp-dmz, describing deployment scenarios and operational considerations for the community. She is also a major contributor to RFC 9832, BGP Classful Transport Planes (IETF Experimental, September 2025), where she is recognized as a co-author in the Contributors section.
An active participant in the IETF IDR and BESS working groups, Reshma brings the combined perspective of standards author, protocol implementer, and network practitioner to this presentation.
Session
At Internet Exchange Points, equal-cost multipath is the norm — but equal-cost is rarely equal capacity. An IXP member with a 10G and a 100G port to the same route server, or a network with multiple upstream exit links of 800G, 400G, and 200G, gets traffic distributed evenly across all paths regardless of what each can actually carry. The result is chronically oversubscribed low-capacity paths and underutilized high-capacity ones. This problem is well understood operationally, but until recently lacked a standardized and interoperable BGP-native mechanism to express it.
RFC 10005, "BGP Link Bandwidth Extended Community," published as an Internet Standard in June 2026, addresses this directly: a BGP extended community encoding link bandwidth in bytes per second that receiving speakers use to compute proportional, weighted load-balancing across ECMP next hops. Its applicability is broad — from asymmetric IXP port configurations and multi-homed peering across exchanges of different capacity, to content providers with WAN uplinks of unequal speeds.
This talk examines RFC 10005 alongside its companion use-cases document (draft-ietf-bess-ebgp-dmz) from both the protocol and operator perspective. We will cover the encoding format, the transitive versus non-transitive community variants and their significance at IXP route servers and peering boundaries, and re-advertisement procedures when next hops change. We will walk through production implementation features — including auto-sensed link bandwidth and aggregate-bandwidth computation across ECMP paths — showing how these reduce manual configuration overhead at scale. The community has also been implemented in production routing platforms, providing concrete validation of deployment readiness. We will also address the peering policy question of when to strip the community to avoid leaking internal capacity information beyond the intended administrative domain.
The talk will also aim to include a customer perspective, with real-world deployment experiences from operators being sought for the final presentation.