[nfsv4]

74th IETF
March 22-27, 2009, San Francisco, CA, USA

WEDNESDAY, March 25, 2009
0900-1015 Morning Session I

=== Agenda Bash and Note Well

Black addition on pNFS and block protocol.

=== Charters

Up on ietf web site.

=== Document Update...

BUTTONS draft on XXX latching - Nico says cleared, and should
clear discusses for that and release all documents...

	Lars...

Editor will wait 

NFS V4.2 spec - says beepy - will be a diff on 4.1, not a whole
new document. Reasserting the decision made when we went with the 4.1
rewrite and reissue approach.

=== Black - Layout Hints for pNFS and bkocks

Implementers hit a layout hint attribute issue.

Specify hint attribute for objects to say how it is striped.
Blocks want to say max timeout value.

Problem - client max I/O time is per client... Dev engineers wants
per filesystem. And there is a fencing check somewhere in here...

Set lease timeouts in the blocks layout draft - minimize text tweaks
to avoid overall delay of publication due to interlocks of drafts 
in the RFC Editor queue (Shepler). 

	There is a fix to be done. Shepler preference is
	to not change anything...

Black wants to add/verify text on lease timeouts on the blocks
draft.

Must be set on every open - timeout - on layout get (Black), if
not set use most rexent value seen by client.

	1. Fix draft to be correct on granularity of attribute.

	2. Need WG alias for any proposed update to spec. Brent
	thinks it is related to lease time set in protocol - with
	client setting shorter time to override in lease time.

Lars - we have time to overlap with Nico's text tweaks...

	Name draft to hold is block layout....

	April 3 draft text release on item two, one week discussion.
	Davic Black to drive.

=== FedFS - Lentini and Tewari

Open portable standards for cross platform federated name space.

Slides are descriptive,

beepy and Brent were touching on implementation of junctions i
prior work like AFS... 

Question: Is a Federation supposed to be in a single administrative
domain. Answer: user name and group to map consistently across
the Federation. Have begun to look at two different realms...

	Crisply define how the docs use the term "federation",
	following discussion of multi-realm stuff.

	Also, consider using different term for "federation".

Renu presented NSDB thoughts and LDAP direction. 

A huge benefit to live without transactions for LDAP. (who
was at mic). 

	Replica updates, and consequences - LDAP change propoagation
	with user consequences. If you need replication semantics,
	(covered on mailing list) you may have problems with
	std. Lentini added text on loose transaction consistency.

	Whether you have to touch more than one entry?

beepy - consider migration hinting to support relaxing constraints
on LDAP usage.

James on security implications. No new security, uses existing one.
NSDB must be as secure as file server - to prevent referral attacks.

=== Labelled NFS

Jared, and Staubach to move forward on this.

Nico does not have a lot of time to spend on this, Nico posted
RPCSEC GSS 3 - will co-author.

If the working group does this, will the vendors implement this.

If we start this, will we finish it.

	Will we get these cycles to move forward.

Dave Black... 

Quigley said we have to move away from MLS systems, and want
to move to SE Linux...

Assessment on amount of work to be done on the protocols? Can
be optional for use in cases like SE Linux.

	Optional recommended attribute...

MAC implementation stuff...

Sun is interested - Jared. More work to do. 

We don't have Labeling expertise here. Though opaque - do we
need sub-structure in to the opaque object...

	Impact, use cases, scope.

beepy to talk to David Quigley about this.

=== Sam on 4.2

umask work - no big objection.

Recall semantics vs. server notification of impending cycling.

Read-ahead bit in read...

Dave Quigley - generic file handle revocation... If a label
changes, allow a generic persistent file handle to be revoked.

	Is there a "data" aspect to some attributes
	requiring a file handle to change?

	Side effects of metdata changes expressed in streams
	of a file.

=== Eisler on sub-file consistency

Coherency vs. consistency...

Brent brought up scientific applications - strided unaligned writes.

Mike wants a compact representation. Brent says think non-overlapping,
vs. power of 2.

Mike presented a couple internet drafts around this work preented 
in Minnesota.

David Black thinks the non-overlapping moving window deduplication
support...

Sub-file coherency...