140 likes | 285 Views
IPoIB. Vivek Kashyap kashyapv@us.ibm.com. IPoIB Draft Status. IP over InfiniBand Architecture draft-ietf-ipoib-architecture-04.txt Approved to be published as an Informational RFC. Transmission of IP over InfiniBand draft-ietf-ipoib-over-infiniband-06.txt
E N D
IPoIB Vivek Kashyap kashyapv@us.ibm.com
IP over InfiniBand Architecture • draft-ietf-ipoib-architecture-04.txt • Approved to be published as an Informational RFC. • Transmission of IP over InfiniBand • draft-ietf-ipoib-over-infiniband-06.txt • Issue: EUI64 -> GUID -> Interface Identifier • Sense of ‘u’ bit • Awaiting IB version 1.2 specification • Resubmit for ‘last-call’ • DHCP over InfiniBand • draft-ietf-ipoib-dhcp-over-infiniband-06.txt • IESG Review
IP over InfiniBand Modes • Unreliable Datagram • Connected • Reliable • Unreliable • Raw
IPoIB Connected Mode • UC • 2^31 bytes • 2-4K links • Yes • No • RDMA Write • RC • 2^31 bytes • 2-4K links • Yes • Yes • Yes Large MTU Automatic Path Migration Reliability RDMA
IPoIB Subnet • IPoIB-Connected subnet is formed over a Non-Broadcast Multiple Access network • IPoIB-UD is over an abstract broadcast link Node A A IPaIPp IPq IPb NodeB NodeD IPd IPs IPr IPc NodeC IB switchednetwork IPoIB UD broadcast domain IPoIB Connected network
Address Resolution IPb IPa • IPoIB UD broadcast/multicast domain • - IPb broadcasts query • - IPa responds UD unicast • IB switchednetwork • - CM REQ message • - CM REP message • - CM RTU message • IPoIB Connected network • - IP communication over connected mode
Address Resolution • IPb: Who Has IPa? • Address resolution over a UD QP • May reuse IPoIB-UD QP • Use multicast-GID • FF1x:<401B>:<P_Key>::<255.255.255.255 or 1> • FF1x:<601B>:<P_Key>::<solicited node address> • P_Key of connected network • [RC|UC]:<QPN>:GIDb • RC|UC indicate the requesters view of the subnet • IPa receives the packet • Determine PathRecord to GIDb • Send reply over UD-QP with GIDa • [RC|UC]:<QPN>:GIDa • The UD QPN is used to create the Service ID for setting IB connection
Connection Setup • Connection is setup using IB MADs • Service ID • Private Data field • Desired MTU • Minimum/Accepted MTU 0x1 Resvd Resvd Resvd Resvd 3-octet QPN
MTU Negotiation per IB link • A requests MTU in private data of CM REQ message to peer B • Desired MTU in private data • B accepts value in CM REP message to peer A • Accepts value in the range ‘Minimum MTU’ and ‘Desired MTU’ • Respond with ‘Accepted MTU’ • A accepts or, • Rejects and tries again Allows a fallback to ‘Minimum MTU’ which is the MTU derived from Broadcast-GID’s MTU
Encapsulation • IP packets encapsulated with the ‘Ethertype’ indication as in IPoIB-UD • Address resolution over UD QP needs to follow IPoIB-UD encapsulation
Ongoing discussions • Multiple connected mode links between peers • Draft says ‘SHOULD NOT’ • Proposal: • Add more detail to draft • Multiple IB connects are allowed. But ‘user beware’. • Mixed UC/RC interconnects • Link type determined based on advertisement • Issue: interoperability
Questions? Vivek Kashyap kashyapv@us.ibm.com