[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[rddp] DRAFT Minn. minutes
Send corrections to me and/or the list. With the
exception of the administrative/procedural matters
on the iSER over InfiniBand draft, all of the decisions
here reflect the sense of the room in Minneapolis, and
may be objected to or commented upon on the list. In
the absence of objection, the sense of the room in
Minneapolis will be the rough consensus of the IPS WG.
Thanks,
--David
The IP Storage (ips) WG met 0900-1130 on Tuesday,
March 8 at the IETF meetings in Minneapolis, MN.
MINUTES (DRAFT)
---------------
Administrivia, agenda bashing, draft status review, etc.: 15 min
David L. Black, EMC (co-chair)
Blue sheets
Note Well
Milestones
Out of date on web site. Update discussion postponed to
end of meeting.
Draft status
All non-MIB drafts except iSER and DA are RFCs or in
RFC Editor's queue.
Elizabeth Rodriguez (co-chair) continues to work with
authors on resolving expert review comments on remaining
MIBs. FC Management MIB has finally made it through this
process, new versions of iSCSI and iSCSI Authorization MIBs
coming soon. iSNS MIB has expired from Internet Draft
servers,
new version expected shortly.
iSER and DA status discussion postponed to end of meeting.
iSER and DA: 45min Mike Ko, IBM
(draft-ietf-ips-iser-01.txt)
(draft-ietf-ips-iwarp-da-01.txt)
iSER = iSCSI Extensions for RDMA
DA = Datamover Architecture for iSCSI
No open technical issues on DA draft - it's ready for WG Last Call.
The open issues on the iSER draft centered on the new
MaxOutstandingUnexpectedPDUs key. The key needs to be specified
so that if the sender violates it (sends too many Unexpected PDUs),
the receiver is *allowed* to drop the connection, but is *not
required* to drop it.
There was a long discussion on when an unsolicited NOP can be
considered "retired" and its "Unexpected PDU" credit can be
safely reused by the sender. Pat Thaler will send detailed text
to specify this to the list.
The draft needs to add advice to implementers on how to deal with
potentially tight target limits on unexpected immediate commands
- the basic idea is to send non-immediate commands, which aren't
subject to the limit, and can cause some preceding immediate
commands
to be considered "retired".
The details of the specification of the MaxOutstandingUnexpectedPDUs
key will be:
Default: "None" (4 letter text string, indicating no limit)
Minimum allowed value: 2
Maximum allowed value: 2^32 - 1
Section 8 of the iSER contains some considerable changes for which
the details matter - WG members are asked to review it carefully.
The X# syntax will not be used with keys added by iSER - they
will be specified by the iSER draft when it becomes a Proposed
Standard
RFC (as a modification of the iSCSI RFC, 3720), hence IANA does
not need to register the new iSER keys, and they should not be
described as "extension keys".
Schedule discussion on these drafts deferred to after next agenda
item.
iSER over InfiniBand: up to 1 hour 30min John Hufferd, IBM
draft-hufferd-ips-iser-sctp-ib-00.{txt,pdf}
This draft is a proposal to generalize iSER to non-TCP RDMA
transports.
There are no changes to iSER over TCP.
The draft requests several changes:
1) Generalize terms/wording in iSER to allow non-TCP RDMA transports
such as RDDP/SCTP and InfiniBand's RDMA service (with RC).
This includes a redefinition of iWARP to encompass SCTP.
2) Generalize wording in iSER to allow a transport to start in
native RDMA mode (with Sends for messages) as opposed to
TCP starting in Stream mode and switching to the RDDP native
RDMA mode.
3) Add some sections on how InfiniBand RDMA works as an example.
4) Extend iSCSI discovery mechanisms to support different
transports.
5) Exempt non-IP transports (e.g., InfiniBand) from "MUST implement
IPsec" requirements.
There were a number of administrative/procedural matters raised by
these requests that were dealt with the WG co-chair (in consultation
with the Area Director (Allison Mankin) in some cases:
a) Item 5) was rejected - the IETF will not approve a blanket
exemption
of usage of a protocol from security requirements. The right
approach
is to refer to RFC 3723 for the security concerns that apply to
iSCSI,
and draft text to require that they are addressed as appropriate in
different transport environments.
b) The authors of this draft have no plans for a draft on iSCSI over
SCTP without iSER. Absent such a draft, iSCSI/iSER/SCTP cannot
be specified, and hence should be removed from the proposal. NB:
subsequent list discussion has indicated possible interest in
writing an iSCSI over SCTP (without iSER) draft, which would make
it possible to specify iSCSI/iSER/SCTP.
c) Infiniband-specific issues, such as dealing with possible lack of
ZBTO support should be dealt with in the InfiniBand Trade
Association,
not the IETF.
d) The RDMAP and DDP drafts have passed WG Last Call in the RDDP WG
with a definition of "iWARP" that is TCP-only (does not include
SCTP).
The usage of the term "iWARP" in this (ips) WG must respect that
usage in the RDDP WG, and hence generalizing "iWARP" beyond TCP is
not appropriate.
At this point, discussion proceeded to the main issue - whether
rough consensus exists in the IPS WG to change the iSER draft to
accommodate to-be-specified usage of iSER over InfiniBand. Making
these changes will likely result in delaying iSER while the details
of the expanded support (e.g., protocol selection information in
discovery) are worked out.
After the discussion, based on a show of hands in the room, the
WG co-chair running the meeting determined that rough consensus
to make these changes does not exist, and hence the iSER draft will
proceed to WG Last Call without any changes proposed in this draft.
During WG Last Call, it will be possible to re-raise these proposed
changes as WG Last Call comments for further discussion.
Given this situation, all InfiniBand-specific material for iSER
should be submitted as a separate individual submission draft (or
multiple individual submission drafts) that make changes to (update)
the main iSER draft and the iSCSI discovery mechanism drafts/RFCs as
necessary. Whether and what of these proposals to adopt as
official
IPS WG work items will be considered at the Paris meeting in early
August.
Based on this, the planned schedule is to issue a WG Last Call for
the DA and iSER drafts in April - authors should prepare versions
ready for WG Last Call by April 15 (tax day), and the WG Last Call
will follow the conclusion of the imminent WG Last Call in the RDDP
WG for the remaining drafts there.
The IPS WG milestones have been accordingly revised to:
Jul 05 Submit iSER (iSCSI Extensions for RDMA) and DA
(Datamover Architecture) drafts to IESG
Aug 05 Submit all remaining MIB drafts to IESG
Sep 05 Review with ADs what (if any) additional work the
WG should undertake
In other words, the intent is to complete the iSER and DA drafts
on the mailing list before the Paris meeting (first week of August).
The Paris meeting will be used to resolve any final MIB issues and
discuss proposed InfiniBand and SCTP extensions to iSCSI and iSER,
with charter revision to follow (Sep) if any of these extensions
are added to the IP Storage (ips) WG's program of work.
----------------------------------------------------
David L. Black, Senior Technologist
EMC Corporation, 176 South St., Hopkinton, MA 01748
+1 (508) 293-7953 FAX: +1 (508) 293-7786
black_david at emc.com Mobile: +1 (978) 394-7754
----------------------------------------------------
_______________________________________________
rddp mailing list
rddp at ietf.org
https://www1.ietf.org/mailman/listinfo/rddp