SSSWG - December 1992 Minutes
IEEE Storage Systems Standards Working Group Meeting Minutes
IBM Boca Raton Working Group Meeting
8-11 December 1992
Over the course of the four-day meeting the following members attended
(thanks to Dave Tweten for supplying this list):
Anthony O. Abbott, IBM FSC Houston, abbott@clearlake.ibm.com
Janet Anglin, International Business Machines Corp., anglinj@stlvm4.vnet.ibm.com
Charles J. Antonelli, The University of Michigan, cja@umich.edu
Betty Jo Armstead, SVERDRUP Tech, Inc., xxbja@atlas.lerc.nasa.gov
Bob Baird, Hewlett-Packard, bbaird@nsa.hp.com
Peter R. Berard, Battelle, Pacific Northwest Labs., d3g349@professor.pnl.gov
Greg Brandeau, Mountain Network Solutions, 71075.420@CompuServe.COM
A. Lester Buck, Compact Data, Incorporated, buck@clearlake.ibm.com, buck@siswat.hou.tx.us
Bob Coyne, IBM, Federal Systems Company, coyne@houvmscc.vnet.ibm.com
John Disbrow, International Business Machines Corp., disbrow@sjevm13.vnet.ibm.com
George M. Ericson, Epoch Systems, Incorporated, gericson@epoch.com
Jim Falgout, Convex Computer Corporation, falgout@convex.com
Richard Garrison, General Electric, no email address
Roger Giudici, Cimage, rug@cimage.com
Andrew Hanushevsky, Cornell University, ABH@CORNELLA.cit.cornell.edu
Ralph H. Herring, E-Systems, emass!ralphh@uunet.uu.net
Harry Hulen, IBM Federal Systems Company, hulen@vnet.ibm.com
D. Creig Humes, NASA Langley Research Center, d.c.humes@larc.nasa.gov
Wayne Hurlbert, Lawrence Berkeley Lab, wayneh@lbl.gov
Bob Hyer, IBM Federal Sector Division, bob@nersc.gov
David Isaac, Mitre, disaac@mitre.org
Tom Jefferson, Sandia National Laboratories, jefferso@sandia.llnl.gov
Peter C. Lawthers, Distributed Computing Solutions, lawthers@discos.com
Steve Louis, National Energy Res. Supercomputer Ctr., louis@nersc.gov
Dennis Luck, Department of Defense, drluck@afterlife.ncsc.mil
Michael S. Milillo, E-Systems Inc., Garland Division, emass!mikem@uunet.uu.net
Donald E. Morgan, Sony Corporation, morgan@sfc.sony.com
Bernie Perella, Department of Defense, bperella@afterlife.ncsc.mil
Ed Poole, IBM Corporation, POOLE@houvmscc.iinus1.ibm.com
Steve Prahst, NASA Lewis Research Center, prahst@lerc.nasa.gov
Rich Ruef, IBM/FSC, ruef1@llnl.gov
Patric Savage, Shell Development Corporation, savage@shell.com
David E. Skinner, Storage Technology Corporation, Dave_Skinner@stortek.com
John Stephens, Severn Companies, Incorporated, stephens@access.digex.com
Dave Tweten, NASA Ames Research Center, tweten@nas.nasa.gov
Joseph A. Wishner, StorageTek, Joseph_Wishner@stortek.com
Richard Wrenn, Digital Equipment Corporation, wrenn@cookie.enet.dec.com
Nancy Yeager, National Ctr. for Supercomputing Apps., nyeager@ncsa.uiuc.edu
Fred Zeiler, Odetics, fredz@odetics.com
George Zerdian, Hughes Aircraft, george@eos.hac.com
----- General Session
Bob Coyne called the meeting to order at 9:14 AM on Tuesday 8 December.
.. SSSWG Sponsor Replacement
As the first order of business Bob noted that our sponsor Pat
Savage is retiring. A smooth transition is assured: Pat will
serve as sponsor until the April 1993 Mass Storage Symposium, and
will continue to be involved with us for as long as it takes to
select and train a successor.
Pat described how new sponsors are selected. The sponsor represents
our working group to the IEEE SAB and other groups, attends a lot
of meetings in addition to our working group meetings, and is
expected to review every PAR that's submitted to NESCOM to check
for overlap. In our charter, the sponsor provides liaison with us,
but we've discovered that such low-level liaison is not what the
sponsor ought to do. The sponsor's job is to make sure that the
way we carry on our work follows all the IEEE rules and that fairness
is always present. Basically, it's a job that doesn't require
someone who is deeply involved in the process; the sponsor is
"behind the front lines." The job requires 1/8 time and quite a
bit of travel, sitting though a lot of meetings, getting a lot of
mail. It also represents a position that must be more conservative
than that of the working group. Our working group wants to move
rapidly, perhaps more rapidly than the current sponsor thinks we
should. The sponsor should be able to restrain the working group
once in a while.
Pat has one name in nomination and is seeking more from the working
group.
Bob Coyne pointed out that this description fits anyone in the room
but him.
When asked, Pat said that he and the rest of the MSSTC Executive
Committee is in total control of the sponsor selection process.
Pat said that this group would not have any input to this process.
Bob asked for a show of hands of those in the working group who
felt that we should have some input into the selection process.
The results of the vote: 21 members felt we should have some input
and 9 members didn't care.
Both Sam Coleman and Bernie O'Lear were approached and both declined
to be nominated. Bob thought Frank Thames from Langley would be a good
sponsor and said we should have more than one candidate to choose
from.
.. IEEE International Standards Effort Financial Support Request
The next order of business concerned a letter sent by the IEEE
regarding financial support of international standards efforts.
Pat asked us to consider what he was going to say simply as input
to the group at this point.
Pat read a latter from the Director of Technical Programs, IEEE
Standards Department. Due to the increasing costs of international
standardization, ANSI wants to charge a fee to standards bodies
such as ourselves. The IEEE has already agreed to this. The fee
will be $300 per year per working group meeting attendee.
Pat then read a reply letter from POSIX, that had been similarly
informed about these new fees. POSIX has declined to pay these
fees for several reasons, all dealing with their perceived inequity
in the fee allocation by ANSI (e.g. targeting only people who come
to meetings is unfair, and those groups that have a fee collecting
mechanism in place will ante up more easily that those groups
without such a mechanism). As a counterproposal, POSIX wants to
add $0.01 per page they ship to their membership and use the
generated revenues to pay the new ANSI fee.
In the ensuing discussion, Pat said he believes the people who come
to meetings should be charged less, since they are already spending
money to come to the meetings, Pat said we should come up with a
more equitable plan, like POSIX has done, for collection of fees.
He suggested $100 per person on the reflector, for example.
It was revealed that several large companies that used to provide
the funding for ANSI international standardization activities have
pulled out, and now the funding must come from alternative sources
such as working group members.
Dave Tweten postulated that since ANSI wants to collect this money
to support international standards activities, it seems that if we
plan explicitly not to have an international standard can we escape
the new fee. Bob Coyne said no, he had tried that strategy already.
Lester Buck asked if the working group had calculated what our fair
share should be. Bob Coyne said Tom Jefferson can figure it out.
Tom Jefferson asserted we do not use more ANSI resources by
having more members attend meetings. Pat agreed and reiterated
that a fee scheme that is not based of meeting attendance is more
equitable.
Dave Isaac said that this seems to be backwards: usually you don't
charge suppliers of a product, you charge the consumers. There
was some general agreement with Dave's position. Dave then asked
how many copies of the RM we distribute per year, and Bob Coyne
replied that in April we were going to freely distribute the RM along
with the proceedings, but that he thought it's feasible to charge
$25 or $50 per copy.
Rich Wrenn then pointed out that that won't work, people won't pay
that, and you can't stop people from making copies. Our goal should
be to get the RM in front of as many people as possible, not generate
revenue from it. I get lots of requests for the RM every week;
the best thing to do is make a copy and ship it to them, not have
them call Bob Coyne and send him $25. This complicates things.
Joe Wishner agreed.
Bob said he was willing to take back the group position on this to
IEEE, whatever it turns out to be, but that we need a few people
to go off and figure out what we think we fairly owe and how we
propose to pay it.
Pat interjected that any fairness arguments this group can come up
with won't cut any ice with the IEEE. This body will be charged
this fee, period.
George Ericson noted that since the fee assessments will start on
1 January, we will need to figure out something soon, but not now,
and moved to table the discussion. The motion was not seconded.
In the continuing discussion the following points were made:
spread out the fee payments, e.g. keep each payment less than $100;
a $300 per person per dot fee already exists in X3; POSIX is IEEE,
POSIX is not X3; for government people it's easier to attach the
new payments onto an existing fee than to create a new fee.
Pat Savage proposed that we tax the people on the reflector. Dave
Tweten replied that NASA may not like IEEE charging for mailings
that come from a NASA machine; the reflector may have to be moved.
Bob Coyne put Tom Jefferson in charge of drafting a response. Dave
Skinner will work with Tom. Anyone with input is encouraged to
get in touch with Tom or Dave. The discussion was tabled.
Dave Tweten said he had paper copies of the reflector traffic for
the past two months available for those who want them.
We next took care of some minor business; we voted on group dinner
selections, and the treasurer called for people to pay their fees.
.. Draft Version 5
Bob Coyne stated that in his view Version 5 is still a draft because
we have not solicited input from industry, and asserted we need to
solicit such input, perhaps via an email collector of some kind.
Dave Tweten said that was a fine idea, since all the wisdom on the
topic isn't in this group. Email may not be the best approach
since some companies are still wary of email. We should set up a
POB too.
George Zerdian asked who should be able to evaluate our work. Bob
Coyne responded that anyone can comment, but we will assign a weight
based on a respondent's field of expertise and degree of involvement.
A discussion followed regarding whether or not the document to be
released at the Monterey Symposium should have the word DRAFT on
it or not. Several points were raised: a draft connotes something
to a limited audience; we should be publishing an official document;
if we don't publish V5, or publish it as a draft, we will never finish
it. Pat asserted that it is the MSSTC who is publishing our document
and therefore has a say in how the document is published. Bob Coyne
asserted that our group can publish V5 as an informative document
at any time without input from the MSSTC. Bob asked for a straw
vote on the question "How many people want us to publish the official
version in April and solicit comments?"; the vote was in the
affirmative by acclamation.
.. Grandfathering voting members
Bob Coyne reminded the membership that Joe Wishner proposed some
voting rights legislation during the previous meeting. In brief,
someone has to have attended two of the last six meetings in order
to vote; Tom Jefferson determines who attended via who submitted
a fee payment. Later in the previous meeting we grandfathered in
everyone present. IEEE said we didn't need to do that, so as it
stands now we have NOT grandfathered anyone in.
This means: in order to vote, a member must have attended three
of the last seven meetings, including the current one.
Pat Savage pointed out that many people are active on the reflector
but don't come to meetings and said they should be able to vote.
No one else could think of anyone that fitted into that category.
The was some discussion of what constitutes presence at an SSSWG
meeting. Dave Tweten remembered that the determining factor is
who checks off the roster, not who pays the fee. The membership
agreed.
Dave wanted to have the voting status appear on the roster, so
people can check their status. A show of hands indicated Dave
should implement this.
Some more discussion on grandfathering. As a result of the
discussion we decided not to change the status quo, that is,
we will NOT grandfather.
Pat Savage said our procedures are usually OK, but whenever something
serious comes up we should follow Robert's rules. Bob Coyne will ask
Dave Tweten to chair in these instances.
.. Quorum
27 voting members were present for the following votes.
Right now we have no quorum, and we take votes regardless of how
many members are present. It was pointed out that we need to have
a quorum for important business.
Bob Hyer moved that a quorum consist of 50% of the voting members.
The motion was seconded.
Discussion.
Rich Wrenn pointed out a problem with our algorithm: people
retain voting status for nearly a year after they drop out, so
it will be hard to muster a quorum under these circumstances,
unless we define a quorum of less than 50%, and Rich doesn't
support that. Rich proposed a 2/3's quorum and a change to the
voting rights to allow a member who has attended two out of the
last three meetings to vote.
With a motion on the floor Rich's motion was not admitted.
Don Morgan suggested three voting levels: show of hands vote,
roll call vote, written ballot vote. He suggested the quorum
be defined in terms of the importance of what we're voting on.
This leads to the notion that important votes be published
ahead of time.
Bob Baird said the original person making the motion can amend it
if the seconder agrees. Pat Savage then said the vote must be
immediate with no discussion.
Bob Hyer amended his motion. The amendment is "I amend the original
motion such that a quorum be 30% of the voting membership".
Immediate vote required. Vote: 21 yea, 4 nay, 2 abstentions.
Amendment passed.
Further discussion on Bob Hyer's motion.
A discussion about last meetings procedures followed. People
who have to leave on Friday can't participate fully. Important
issues need to be pre-announced.
Vote on Bob H's motion: 18 yea, 7 nay, 2 abstentions.
Motion passed.
Dave Tweten moved that we modify the definition of a voting member
to permit someone to resign as a voting member and have his/her
name removed from the list for quorum purposes.
Discussion.
Lester Buck saw no incentive for people to resign. Dave Tweten
refuted him by citing three people who did that. Tom Jefferson
reminded Dave that one of the three resigned before we had a
formal concept of a voting member.
George Ericson said we should not forget who was where when.
George Zerdian asked Dave Tweten if he would accept an amendment to
his motion that anyone wishing to resign as a voting member must
send hard copy to Dave and that Dave retain this hard copy.
Dave accepted the amendment; the amendment was seconded.
Vote: 5 aye, 16 nay, 6 abstentions.
Amendment defeated.
Vote on Dave Tweten's motion: 22 aye, 0 nay, 5 abstentions.
Motion passed.
.. Consensus
Bob Hyer moved that consensus among the membership be defined as
50% plus 1 vote of the attending eligible voting members.
The motion was seconded.
Discussion.
Rich Wrenn never heard of a consensus figure that low. and proposed
that consensus consist of 85% of attending eligible voting members.
The motion was not seconded.
It was pointed out that whenever we take a vote, we usually go by
50% + 1, i.e. simple majority.
Bob Hyer withdrew his motion.
George Ericson moved that we operate under simple majority with the
following exceptions, allowing the membership to raise the level
of consensus, by simple majority, to: consensus of voting members
present and total voting membership, consensus being defined as
75% of either group. Seconded.
Discussion.
The floor proposed an amendment that the definition of consensus
for total voting membership be 100%. The motion to amend was
not seconded.
Bob Coyne named Dave Tweten as vice chair. The vice chair immediately
recognized Bob Coyne.
Bob proposed an amendment to impose a time limit between the time
something is voted on at a meeting and the time all votes need to
be registered; failure to reply within a time set by the membership
shall count as an abstention. The motion to amend George Ericson's
motion was seconded.
Vote: 10 aye, 7 nay, 12 abstentions.
Amendment passed.
Dave Tweten said George's motion should be withdrawn since this sort
of thing should be put in the bylaws (if the objective is to avoid
the majority trampling the rights of the minority, this motion cannot
prevent that).
Dave returned the chair to Bob Coyne.
Further discussion of George's motion.
A motion was made to table further discussion of George's motion.
Motion to table was seconded.
Vote: 24 aye, 1 nay, 2 abstentions.
Motion to table passes.
Bob Coyne suggested that we revisit this topic after the issue has been
investigated further by the interested membership. It was suggested
that Bob put the bylaws on the reflector.
.. Upcoming Meetings
NCAR Feb 9-12
Monterey 25-30 April (Half of Thursday and all day
Friday)
LaRC 22-25 June
DEC Colo Spgs 10-13 Aug
NASA LeRC 12-15 Oct
Bob Coyne said anyone wanting to organize a ski trip for the NCAR
meeting should do so. We should be reserving rooms for NCAR now,
or we might not be able to get them.
.. New Business
Our list of PARs now is:
Existing:
1244.1 SSS.OID Standard Object Identifier for Storage Systems
New:
1244.2 SSS.PVL Standard Physical Volume Library for Storage Systems
1244.3 SSS.PVR Standard Physical Volume Repository for Storage Systems
1244.4 SSS.MVR Standard Data Mover for Storage Systems
Pat Savage reported that NESCOM said our PARs are very good models
of what a good PAR should be.
Pat said we should structure ourselves move like POSIX, e.g. we need
an technical executive committee chaired by Bob whose members are
the [so-called] dot chairs. Bob Coyne agreed.
.. Technical Editing
After a distribution of meeting room assignments, Dave Isaac gave
technical editing instructions.
1. Review modules you use for services you need
a. consistent
b. complete
c. concise
The review list (X -> Y means subcommittee X reviews subcommittee
Y's section of the document)
Must do:
BFS -> SS
SS -> Mover
Mover -> PVL
PVL -> PVR
PVR -> NS
NS -> SSM
SSM -> BFS
Should do:
BFS -> Mover
SS -> PVL
PVL-> Mover
2. Deliver one marked-up copy of document to the reviewed
subcommittee, and one set of issues to Dave.
(Due by COB tomorrow 4:30-5:30)
3. Thursday
a. Review another module
b. Review changes to you
c. Issues review at 3pm
4. End of week
One changed hard copy or soft copy per subcommittee.
5. Changes
a. Technically substantive (use change proposal form)
b. Editorial (redline the hardcopy, attach sheets if necessary)
Bob Coyne volunteered the IBM art department to draw the pictures
for the document in a consistent style.
Dave Tweten reported that he and George Erickson ran into the problem
that Word for the Macintosh and Word for the PC don't overlap. Dave
Isaac asked everyone to fax him any pictures, since he'll redo them
all in Canvas anyway.
Bob Coyne adjourned the meeting at 12:00 Noon.
.. Voting Summary
1. A quorum is 30% of voting members.
2. Anyone who wants to drop his/her membership can do so by notifying
Dave Tweten (hard copy not necessary).
----- General Session
Bob Coyne called the meeting to order at 2:12 PM on Thursday 10 December.
At roll call we mustered 26 members present.
.. Financial Report
Tom Jefferson gave a financial report. It costs about $1000-2000 to
hold a meeting.
Our financial balance sheet:
$4500 in account before this meeting
750 required minimum balance
2000 to prepay next meeting's expense
-----
1750 cushion
Tom announced that the bank (Homestead, Livermore CA) in which we
are keeping our working group funds has been seized by federal
authorities. In consultation with working group officers Tom has
switched our funds into Great Western Bank, Livermore CA. In the
new bank, we have two accounts, both a savings and a checking
account. Two signatures are required from among the working group
officers (Bob Coyne, Sam Coleman, Charles Antonelli, Tom Jefferson)
to withdraw funds. Tom Jefferson alone is authorized to transfer
funds between the savings and checking accounts.
When Tom put the question if any member objected to these new
arrangements, none did.
.. Model Review
Dave Isaac handed out the modified review agenda. We will start
at the top of the model and work our way down.
-- Editor (Dave Isaac)
1. Title page change.
Accepted.
2. Reorder paragraphs around the Model picture.
Accepted.
3. Fix section numbers.
Accepted.
4. Separate PVR/PVL into sections
Accepted.
-- Architecture (Betty Jo for Mike)
1. Substitute term "component" in lieu of model component.
Rejected, will use component name in all descriptions.
2. Change SSOID to OID and BFID to FID.
Resolution: rejected.
Andy: we have had SSSOID, SSOID, SOID, and OID. I'm confused.
3. Change reference from bitfile to object.
Tabled.
4. Add definition of compliance.
Resolution: decided there should be a section in the document
that says how to use the document.
Discussion of the necessity of including a statement to the effect
that it is not possible to be in compliance with this document
and where to include it. Bob Coyne moved the responsibility for
the content and placement of this section to the technical editors.
5. Varying level of detail in document.
Resolution: should be a limit on the level of detail and that's
the responsibility of the editor.
Discussion of whether or not the model should be describing
optional features. John Stephens points out that it is difficult
to decide what is optional and what is not.
6. Unify naming of components ("bitfile server" => bitfile manager,"
etc.
Dave Isaac: hard to tell what's in the model and what isn't. We
chose "manager" to tell us that things are in the model. For
another thing, we want the names to be more technically correct
than they are now. As examples, the storage server does not serve
storage any more, and the PVR is not a physical volume repository
any more, it stores cartridges which have multiple volumes on them.
Straw poll: (voting by individual model component names):
PVR becomes "Physical Media 'something'": 18 yes 7 no
Objection from the PVR subcommittee to the word "media."
Dave Tweten suggests we should vote for them as a bloc, accompanied
by general agreement.
Straw poll: every part of the model ends in "manager."
yes 12 no 15
Straw poll: every part of the model ends in "component."
yes 12 no 13
Therefore every part of change recommendation 6 has been rejected.
7. Server vs. service difference not defined in glossary.
Accepted.
8. Define media in glossary.
Accepted.
9. Structure and layout of sections should be mutually consistent.
Accepted.
10. Server not defined in glossary.
Accepted.
11. Where does the LS go in Figure 1.
Tabled.
12. "Server" versus "service."
Resolution: Use "server."
13. Mover should delete PVR general object information (e.g. no object
oriented slant in document).
Accepted.
14. Move "Storage System Environment" section to a new Section 6
instead of 5.4 and entitle it "Storage System Management."
Accepted.
15. Move security description to Storage System Management section.
Accepted.
16. Replace octets with bytes.
Accepted.
17. Missing update to section 4 (from Andy)
Accepted.
18. Need consistent capitalization (if it's in the model - capitalize it)
Accepted.
19. Renumber sections.
Accepted.
20. Object identification (insert SOID in text in the right places).
Accepted.
21. New (3D) model picture.
Accepted.
22. Spelling error.
Accepted.
23. New picture from George.
Accepted (#21 handles it).
24. Add client list to each description.
Accepted via restructuring.
25. LS overview in 5.3.
Accepted.
26. NS overview.
Accepted.
27. BFS overview.
Accepted.
28. Mover overview in 5.3.
Accepted subject to possible change.
29. Not enough detail in Storage Server overview.
Tabled (accepted in principle).
30. Model inconsistency in PVL.
Agreed that PVL section is inconsistent, but submitted change
was rejected. Accepted need to rewrite.
31. Model inconsistency in PVR.
Agreed that PVR section is inconsistent, but submitted change
was rejected. Accepted need to rewrite.
32. SSM readability.
Accepted need to rewrite.
33. Typo - duplicate.
34. Two descriptions of the BFC - consolidate.
Accepted.
35. Typo.
Accepted.
36. Remove BFS connection in picture.
Done.
37. Move 2.3.4 to follow the Storage System Management section.
Accepted.
38. Move or delete 3.6 to 4.X. (page 40).
Accepted, delete.
39. Standard Object Identifier should be changed to Storage Object
Identifier.
Andy: either SOID or SSSOID. Your choice.
Tabled.
-- Bitfile Server (Dave Tweten)
1. Remove leveling from bitfile containers.
Accepted.
2. Eliminate BFS service interface.
Accepted.
3. Remove section 3.1.5.1.
Accepted (duplicate of 2).
4. Multiple containers in virtual volumes.
Rejected.
5. Eliminate bitfile containers.
Rejected, but will call the SOID field the "creating bitfile container."
6. Rewrite language referencing "bitfile strings."
Accepted, will eliminate the term "bitfile strings."
7. No longer use "machine-oriented id."
Accepted, the terms "system-oriented id" and "machine-oriented id"
are deprecated, the terms "container SOID" and "object SOID" are to
be used.
8. Use Bitfile regions.
Rejected, bitfile regions are not objects.
9. Command reference introduction should be made less general.
Accepted, reworded.
10. Is the BFS the only place where repacking occurs.
Accepted the question, BFS not the only place (SS can do it on a volume)
11. Remove or move section 3.2.1
Accepted, there will be an Examples section after the Commands
section, and it will be more concise.
12. Remove definition of "holes," seems to be implementation-dependent.
Rejected, this proposal lacked a counterproposal (BFS came up with
five models), want a counterproposal.
Discussion:
Several possibilities:
1. Holes are verboten
2. Unix semantics
3. Unix semantics after a copy
4. Get an exception - partial return
Straw vote - vote is to accept the subcommittee's decision.
13. BFS procedure call interface too specific.
Accepted, last 3 paragraphs of 3.1.5 deleted.
14. Remove redundant BFC sections from BFS.
Accepted, text deleted.
15. Shorten overview for BFS.
Tabled.
16. Move Scope section for BFS.
Accepted.
17. Explain or delete BFS_Sync, BFS_Wait, BFS_Reply
Accepted, will explain more clearly.
18. Detail on content of BFID should be removed.
Accepted.
19. Word "collection" should be changed to "container."
Accepted.
20. Section 3.1.8 should be moved.
Accepted.
21. Multiple IDs in the BFS.
Accepted, BFID is instance of an SOID; bitfile descriptor is an
implementation detail.
-- Break
-- Mapping Services (Andy Hanushevsky)
Received 39 change requests: accepted 49%, conditionally accepted %33,
and rejected 18%. Andy only went through the rejects in the interest
of saving time.
1. Wording change
Rejected.
2. Have a directory service architecture whose particular
instantiations are LS or NS.
Rejected.
3. Condense/reword LS.
Describe where LS database resides. Rejected.
Change phrase "logically single server." Rejected.
Define "well-defined." Rejected.
Straw poll: "how many people want to strike the first sentence of
section 3.2.2". 17 yes votes; sentence will be struck.
4.
Rejected, we already use components.
5. Move general rationale in NS/LS to architecture section or
delete it.
Rejected, need to keep NS/LS rationale in NS/LS section.
6. Take out text describing hints and caching.
Rejected, need to keep them but emphasize they are recommendations.
Discussion, members felt this was too implementation-specific.
7. Define "trusted computing base" in Glossary.
Rejected, it's an IEEE defined term.
Dave Isaac accepted responsibility for the Glossary.
8. Change first sentence of 5.4 to read "The storage system
environment consists of storage system domains that ...
set of other objects."
Reject "other" - precludes a container from containing itself.
Re "domains: (as opposed to (domain); suggest Figure 1 title
be changed to "Environment for a Storage System."
9. Description of Optional items
Dave Tweten: since everything in the model is technically optional,
it's not necessary to distinguish pieces of the model as "optional".
The remaining change requests consist primarily of rewording changes.
-- Storage Server (Lester Buck)
1. The client is responsible for managing the logical address space.
Accepted.
2. Describe necessary PVL objects in SS.
Accepted.
3. Client partition -> physical data space.
Accepted.
4. Remove restriction that you can't update a sequential volume.
Accepted.
5. Should describe features of objects created at a lower level.
Accepted.
6. Miscellaneous rearrangements.
Accepted.
7. Remove term "human-readable string."
Accepted.
8. Remove "base volume" terminology, etc.
Accepted.
9. Use "physical volume" instead of "segment."
Accepted.
10. Should allow simple volume on top of physical data space.
Accepted.
11. Distinct physical and virtual volume sets.
Accepted.
12. Provide capability to write arbitrary end of tape.
Accepted.
Considerable discussion.
13. Miscellaneous edits.
Accepted.
-- Mover (Bob Hyer)
1. Mover should support "load" command.
Accepted.
2. "Transfer complete" message comes from originator only.
Rejected, existing systems need separate client responses for
performance reasons.
3. Wrong verb names.
Rejected based on Dave Isaac's input.
4. Define mover object.
Accepted.
5. Define an option for implicit buffer flush after move complete.
Accepted.
6. Delete interface asynchronous notification directive.
Rejected.
7. "Partition" is too general a term, use another.
Accepted, will use "ambit."
8. Device control commands are incomplete.
Accepted, added get/set attribute.
9. No need for request queue object.
Accepted.
10. Mover from memory.
Accepted.
11. Remove all references to data transformation operations.
Bob Hyer claims we need to retain references only in the cases of
encryption, compression, and error correcting codes.
Considerable discussion. To support, say, encryption you need to
look at the bits, and this limits vendor interoperability. Lester
claims this opens the door to a can of worms. Dave Tweten claims
these operations are themselves interpretations, so they cannot be
described as acting on streams of uninterpreted bits.
Straw poll: "how many people think movers must be allowed to do
encryption and compression." Vote is nearly unanimous in favor.
12. Clarify partition command semantics in Mover.
Rejected.
13. Reword mover as peer-to-peer exchange.
Rejected.
14. Remove interface references from mover.
Accepted.
15. Correct mover diagram errors.
Accepted by correction or deletion.
16. Delete mover examples.
Rejected.
17. Change mover figure 1.
Rejected (the figure has been changed anyway).
Bob showed several mover figures. Some discussion about the figure
contents.
Lester volunteered to draw a new mover figure 1.
-- Storage Systems Management (Steve Louis)
1. Unclear paragraph on p. 50.
Accepted.
2. Paragraph defining packages in management model is not used
in following text.
Accepted, paragraph deleted.
3. Delete "superuser" p.50 line 1761.
Accepted.
4. Section 3.7.1.5: short term performance measurement is
important too.
Accepted.
5. When enumerating lists use nonexhaustive terminology.
Accepted.
6. Change "bitfiles" to resources.
Accepted, will cite as a specific example.
7. Section 3.7.1.3: good security does not build policy into programs.
Accepted, will delete.
8. Put section 3.7.3 into environment section.
Rejected.
Bob Coyne said Steve Louis is going for a PAR and asked if
anyone else wanted to too.
Tomorrow morning we will finish up by reviewing PVR/PVL change
requests. The meeting will start at 8:30.
Meeting adjourned at 5:28.
----- General Session
Dave Isaac called the meeting to order at 9:01 AM on Friday 11 December.
No quorum was taken.
-- PVL (Dave Skinner for Rich Wrenn)
1. Reference MLD/DLD's on p.31.
Accepted.
2. Two-sided media.
Accepted with modifications.
3. Need to extend the functional cartridge concept with a more
general name; e.g. the PVL should not be restricted to removable
tape volumes only, should also be able to handle removable disks.
Tabled, need to come to agreement across entire model with respect
to class inheritance.
Joe Wishner said the PVL subcommittee recognizes the problem and
has no objection to changing the name but claims it is too pervasive
a change to be included in Version 5 of the document. There was
an extended discussion about the issue, during which Dave Tweten
was surprised to learn that the PVL has taken responsibility for
managing non-removable media.
Dave Isaac stated the editor's position on the Glossary: the
Glossary is only for English words that have a special use in our
context; all object descriptions must be in the main text.
Dave Isaac will prepare an index of object definitions.
The PVL will describe an object hierarchy. The following discussion
centered around the topic of the necessity of class hierarchies in
the various other sections of the Model document.
A straw poll was taken regarding the question of who wants to add
class and/or containment hierarchies to the rest of the model.
Voting results: add class: 9, add containment: 10, against either
idea: 2.
Joe Wishner pointed out we don't have enough time to add these
hierarchies. Steve Louis thought we had previously agreed not to
do object hierarchies. The membership agreed that we'd like to add
hierarchies but don't have the time.
Lester Buck volunteered to do the class/containment hierarchies for
the entire document. The membership agreed that if Lester doesn't
build consensus or get the hierarchies to Dave Isaac there won't be
any hierarchies. Results of a straw poll on whether we should let
Lester do this: 8 aye, 3 nay.
4. Remove delete set/assign set functions from PVL.
Rejected, these functions are intended to support multiple physical
volume sets.
Dave Isaac did not accept this resolution and tabled the issue instead.
-- PVR (Joe Wishner)
1. DLD supports multiple media types, staging area should too.
Tabled, doesn't scale well into MLD/DLD.
2. Ports should be able to do both in and out types.
Accepted.
3. Add cartridge to glossary.
Rejected, no objects in glossary.
4. Clarification of definition in p.2.
Accepted, given to Dave Isaac.
5. Modify introductory material.
Accepted, given to Dave Isaac.
6. Mover list attribute omitted from drive object.
Rejected, it's an arch issue whether the PVR needs to store
info the PVL needs but the PVR does not need.
Considerable discussion of whether or not a model object should
store metadata for another model object. Dave Isaac's view is that
metadata should be stored only with the object that owns it, and
other objects that need the metadata will need to access it from
the owning object. Dave Tweten pointed out that any model object can
be a client of the model, and use a bitfile to store its metadata.
This results in a bootstrapping problem, though.
Bob Coyne interjected some other business.
George Michaels and Dick Watson will be asked to help edit, since Sam
Coleman is busy with the Symposium Proceedings until mid-January.
Dave Isaac will keep putting copies of the evolving document on
att.research.com and will make arrangements for paper mailings.
Bob Coyne is looking for show-stopper comments only (e.g. "the document
is nonpublishable because ...") in February, since we will never agree
on everything.
Joe Wishner wants a "things to be done" section for vendor comments.
Bob Coyne reiterated that we should be past the "taking potshots"
stage. Arguments such as "it won't scale" aren't good enough; you
have to show how the model doesn't scale.
Next, we took a straw poll on whether or not we want an Issues Section
in the published document for unresolvable conflicts, concerns, and so
forth.
Voting results: 10 for such a section, 5 against.
--- Break
We returned to the discussion of whether or not a model object should
store metadata for another model object in a different layer.
A comment was made that this isn't metadata but metadata attributes
we're talking about here.
George Ericson asserted that a bitfile is an arbitrary named set of
attributes. It may know about some of those attributes but the BFC
can insert as many other attributes as the BFS implementation will
support. As a general concept the support of some named attributes
is very reasonable.
A comment was made that it looks like we've merged structured and
object views, and this causes the problem: we need to provide a view
of data stored in one layer in another layer.
Dave Tweten called for a straw vote on whether or not a model object
should store metadata for another model object in a different layer.
Discussion.
Dave Isaac said he didn't care which view we take as long as we are
consistent.
Straw Poll: "The PVR does NOT have to define attributes that are
only of interest to the PVL."
Vote results: 16 aye, 1 nay.
Straw Poll: "Do we admit that an object can have an attribute called
'client-specified-information' that the client can set and retrieve
and that the client understands but that the object does not."
Tom Jefferson asked about multiple clients overwriting a single
client-specified-information attribute.
Discussion both for and against allowing overwriting.
George Ericson asserted that a single attribute is not sufficient;
a more general mechanism is to allow arbitrary named attributes.
Now, while the BFS has unlimited storage for arbitrary attributes,
other model components may not. So George suggested the model provide
an interface to each object - called "Get/Set Attributes" - where
an implementation may or may not support an arbitrary number of such
attributes; it is up to a client to work with any particularly
restricted implementation.
Dave Isaac said we have an object oriented view mixed with a server's
view of an object. Dave said it's clear that files have to have some
client-specific information, and thought the level of detail in the
model (name-attribute) is not enough; we need to specify more
information about the attributes than that. Dave thought the PVL
should create their own object and be done with it.
Lester reminded us that there are two things (PVL/PVR) because we
wanted to separate the functions and asserted that storing data with
objects that don't own the data is breaking the model. A comment
from the floor said that while in general we shouldn't so this
storing, in the specific case of the PVR and PVL we should allow it.
Straw Poll: "If you want client specific information stored in an
object with Get/Set operations defined for access to that information
then vote aye."
Vote results: 2 aye, 12 nay.
Straw Poll: Repeat previous vote for the specific case of PVR/PVL:
client specific information is the mover-id, the storing object is a
drive object located within the PVR object, the accessing object is
the PVL.
Vote results: 4 aye, 9 nay.
We returned to the list of issues.
7. MLD needs containing cartridge type attribute.
Rejected, mapping already provides this information.
Dave Isaac tabled the resulting lengthy discussion.
8. Clarify compatible DLDs attribute.
Accepted.
9. #200 Inconsistency of attribute definitions in specification.
Accepted.
10. #201
Rejected.
11. #202 Improved attribute names.
Accepted.
12. #203 Need physical name attribute for other objects beside
port object.
Rejected, need more information.
13. #204 Change the "move" verb to "relocate."
Tabled. Dave Isaac will try postfixing to uniqify names.
14. #205 Infinity should appear as a common definition.
Accepted by removing reference to infinity.
15. #206 Fix incorrect sentence.
Accepted.
16. #207 The PVL does not support cooperating PVR's.
Tabled.
17. #208
Rejected.
The PVR group still has some issue sheets to evaluate.
Lester asked for the PVR figures.
Dave Isaac said he needs changes by the end of next week.
Meeting adjourned at 11:24.
-----
Respectfully submitted,
Charles J. Antonelli
Secretary, IEEE SSSWG
SSSWG Home