90 likes | 289 Views
iSER Draft Status. draft-ietf-ips-iser-01 Mike Ko March 8, 2005. Declared Limit to Control the Number of Unexpected PDUs.
E N D
iSER Draft Status draft-ietf-ips-iser-01 Mike Ko March 8, 2005
Declared Limit to Control the Number of Unexpected PDUs • A new key, MaxOutstandingUnexpectedPDUs, can be used by both the initiator and the target to declare the maximum number of outstanding “unexpected” control-type PDUs it can receive • For the PDUs listed on slide 3, the PDU is outstanding until the target responds with the corresponding response PDU • NOP-out PDU with ITT = 0xffffffff and TTT /= 0xffffffff (ping echo) is not subject to the declared limit since it is sent as a response to a NOP-in PDU from the target • Similarly, NOP-in PDU with ITT/= 0xffffffff and TTT = 0xffffffff sent as a response to a NOP-out PDU from the initiator is not subject to the limit M. Ko
iSCSI Control-Type PDUs from the Initiator Regulated by CmdSN Window M. Ko
Handling of Unidirectional NOP-out PDUs at the Initiator • For a unidirectional NOP-out PDU from the initiator with ITT = TTT = 0xffffffff, CmdSN is not advanced • To retire a unidirectional NOP-out PDU, needs confirmation from the target that the PDU has been processed • A unidirectional NOP-out PDU is outstanding until the target responds with a control-type PDU with ExpCmdSN set to at least x + 1 where x is the CmdSN of the PDU with the immediate flag set • For a session with multiple connections, a PDU with no immediate flag and CmdSN=x sent in a different connection could trigger a control-type PDU with ExpCmdSN=x+1 from the target • To avoid ambiguity, the control-type PDU from the target with ExpCmdSN=x+1 (or larger) must be received on the same connection as the unidirectional NOP-out PDU in order to retire the NOP-out PDU • An implementation note was also added which recommended that “To avoid complexity, a NOP-out PDU with ITT /= 0xffffffff can be used instead” M. Ko
Handling of Unidirectional NOP-in PDUs at the Target • Similarly, for a NOP-in PDU with ITT = TTT = 0xffffffff and StatSN = x, the PDU is outstanding until the initiator responds with a control-type PDU on the same connection where ExpStatSN is at least x+1 • An implementation note was also added which recommended that “To avoid complexity, a NOP-in PDU with TTT /= 0xffffffff can be used instead” M. Ko
Changes in -01 Version • Section 6.7 was added to describe a new key, MaxOutstandingUnexpectedPDUs, which allows both the initiator and the target to declare the maximum number of outstanding “unexpected” control-type PDUs it can receive • Last paragraph in section 7.3.4 was modified to clarify that SCSI Data-out is not used for solicited data since an R2T is always transformed into an RDMA Read Request • Section 8.3 and 8.4 were added to describe the buffering requirements for the expected and unexpected control-type PDUs and when these PDUs are no longer outstanding • Described the use of the MaxOutstandingUnexpectedPDUs key • Added an implementation note to suggest using NOP-in and NOP-out as ping request and ping response to avoid complexity M. Ko
On MaxOutstandUnexpectedPDUs • Consensus needed for the following parameters: • Default value is “none” • Minimum value is 2 • Rationale: RFC3720 states that “An iSCSI target MUST be able to handle at least one immediate task management command and one immediate non-task-management iSCSI command per connection at any time” • Maximum value is 2**32-1 M. Ko
Things to Do: IANA Considerations • Need to register new keys with IANA • RDMAExtensions • TargetRecvDataSegmentLength • InitiatorRecvDataSegmentLength • MaxOutstandingUnexpectedPDUs • Need to indicate in the IANA Considerations section that these new keys are in the “iSCSI extended key registry” • Need to add reminder in the key descriptions that these keys should be used with the X# prefix • RFC3720 states in section 12.22 that “For IANA registered keys the string following X# must be registered with IANA and the use of the key MUST be described by an informational RFC” • Assume the intent is for an “informational RFC” or better M. Ko
Things to Do: Generalized Support for IB and SCTP? • draft-hufferd-ips-iser-sctp-ib-00.txt described changes/adjustments to be made to the iSER draft to generalize the transport layer to include other RDMA-capable protocol layer • Allows iSCSI/iSER to layer over SCTP version of iWARP, or Infiniband, etc. • Only the iSER draft will be affected if the recommendations in draft-hufferd-ips-iser-sctp-ib-00.txt are accepted • No impact to the DA draft M. Ko