IP over InfiniBand: 55th IETF Atlanta (11/18/02 - Meeting Minutes by Sunay Tripathi) ================================================================== Agenda Charter update IPoIB MIBs status IPoIB connected mode Wrap up Charter Update - Jerry Chu (Sun Microsystems) IPoIB connected mode was added and approved by IESG. Care must be taken to avoid interference between reliability algoriths at different level. See the updated WG web page http://www.ietf.org/html.charters/ipoib-charter.html Drafts under IESG review Link and multicast draft Encapsulation draft DHCP over IB draft ADs have already commented. Need to address ADs comments. MIBs Status - Sean Harnedy (InfiniSwitch corporation). 6 MIBs currently defined in the WG charter Sean discussed the current status of each MIB (what was the current revision etc). Next steps - Interest in defining new MIBS? Update IDs to new 1.1 IBTA sepc. Advance current MIB work. Gather any implementation experience. Any Interest in a IPoIB Management overview doc? Advance certain IPoIB MIB to WG last call. IPoIB connected mode - Vivek Kashyap (IBM) Vivek will mail out the presentation slides to the mailing list. The main point in the presentation was you need to know the destination GID to initiate connection and possibilities on how to resolve the GID. Jerry asked if the proposal uses IP address hosted on a UD QP to identify its equivalent RC/UC QP, wouldn't it require a thousand IB connections when there are a thousand IP addresses hosted on the same UD QP interface? Vivek said he needs to think about this. Jerry mentioned a similar issue with the UD QPN - it takes GID+QPN (UD) to uniquely idetifiy an IPoIB link interface. But UD-QPN is not present in the proposal when identifying RC/UC QP. Some discussions continued on how MTU is determined. AD commented that it's probably worth defining a default MTU for IPoIB UC/RC to save some trouble. A follow-on comment was made on using two separate connections for IPv6 and IPv4 because the former supports jumbograms (> 64KB). Wrap up - Jerry Chu Now that the IBA 1.1 specs have been released, Jerry asked the draft authors to verify the contents of their drafts against the 1.1 specs. AD (Thomas Narten) asked how much more change is expected in the future, e.g. what is the next IBA release (2.0?) and when? Will any future change require modifications to the IPoIB specs? Jerry believed that is unlikely because IBTA must maintain a high degree of compatibility between newer specs and the old ones. The 1.1 release mainly addresses the management and multicast areas, both are incomplete in the 1.0 release. No more questions were asked so the WG meeting was concluded.