An anycast IP address [RFC 1546] is an address assigned to a set of nodes, such that a packet sent to that address is delivered to one member of the set, in particular to the member "nearest" the sender according to the routing system's measure of distance.  Anycast addresses are formally defined in the IPv6 address architecture [RFC 2373], and are also being used or proposed in a number of IPv4 contexts. There are a large number of open issues regarding anycast addresses, listed below.  The purpose of this BOF is to discuss these issues, identify any additional issues, and determine what has to be done to resolve them (e.g. find volunteers to write one or more drafts, perhaps propose establishment of an IETF working group).  The known issues requiring resolution are: - more thorough definition of anycast semantics (what does   "nearest" really mean?  what happens if more than one   nearest member?  is delivery to at-most-one member,   at-least-one member, or something else?), including   defining special cases of anycast addresses (e.g.,   "cluster addresses") - pros and cons of having anycast addresses be syntactically   distinguished from unicast addresses, and of having "well-   known" anycast addresses? - effects on routing (how are they routed?  how can/should   they aggregate?  locally-scoped vs. global anycast addresses?   limits on usage due to routing scaling limits?) - transport protocol issues, e.g., supporting TCP connections   to anycast addresses - allowing hosts to be members of anycast sets -- requires   a host-to-router protocol like IGMP, but strongly   authenticated? - usage guidelines, e.g., anycast vs. multicast for resource   discovery?  local use vs. global use? Attendees should have read the following documents in advance of the BOF: - RFC 1546, Host Anycasting Service - RFC 2373, IP Version 6 Addressing Architecture (section 2.6) - RFC 2526, Reserved IPv6 Subnet Anycast Addresses If you would like to present specific ideas/proposals to the BOF, or if you know of other relevant documents that should be on the pre-reading list, please send email to Steve Deering .