SSSWG - July 1990 Minutes

From: tweten@gilmore.nas.nasa.gov (Dave Tweten)
To: ieee-mss@prandtl.nas.nasa.gov
Subject: IEEESSSWG 7/90 Minutes
Date: Sat 04 Aug 90 21:38:38

The first meeting of the IEEE Storage System Standards Working Group
was called to order on Wednesday, July 25, 1990 by Chairman Sam
Coleman.  Over the two days of the meeting, the following people were
in attendance:


	Bob Burns
	Loellyn Cassell
	Sam Coleman
	David J. Donald
	Dr. Elias Drakopoulos
	Jack H. Franklin
	Bruce Griffing
	Bill Hammond (representing Janet Anglin)
	Andrew Hanushevsky
	Gene Harano
	Michael Hardy
	Harvard Holmes
	Harry Hulen
	Tom Jefferson
	Merritt E. Jones
	David Kitts
	Peter Lawthers
	Steve Louis
	Dennis Luck
	Jun Mao
	Michael S. Milillo
	Tom Nehrig
	Bernard T. O'Lear
	Steve Prahst
	Chuck Pross
	Dennis F. Reed
	Bob Riddle
	Jack J. Rousso
	Patric Savage
	Robert P. Seal
	J. T. Thompson
	Dave Tweten
	Greg Williams
	Jim Williams
	Joel Williams
	Richard Wrenn
	Ken Yang
	George Zerdian

Each attendee gave a brief introduction.

Pat Savage gave a report on the status of the efforts in acquiring a
formal standards number from the IEEE.  He said the original Project
Authorization Request (PAR) had been returned because the IEEE was
changing its procedures.  There is now an issue of one PAR versus
multiple PARs.  Pat will attend the next standards board meeting in
September and hopes to return with a number.

A review of the various forms of standards was also given by Pat.  The
first is a Guide (which may or may not be based on the Reference Model
with changes).  The Guide can also be used to reflect some of the
arguments on how and why certain decisions were made.  The next level
is the Recommended Practices followed by the full Standard.  Hope was
expressed that some effort on all three levels could be done in
parallel.

Sam gave his opening remarks which included emphasis on using the
E-mail reflector set up through NASA-AMES to facilitate discussion on
the various issues which will arise.  He also reminded the attendees
that the proceedings are in the public domain.  Votes taken in the
workshop meetings are informal.  Formal votes will be conducted under
IEEE rules.  Draft standards will be sent out for public review and all
received comments must have a response.

Regarding finances, this is a volunteer working group.  Registration
fees will be used to cover meeting room and refreshment expenses.

Considerable discussion occurred when several of the vendor attendees
mentioned the problems they were having in responding to solicitations
which required compliance with the Mass Storage System Reference Model.
Pat clarified the situation by stating that this working group is not
responsible for that document.  There will be no further releases of
the Reference Model.  Everything in it is subject to change as this
working group evolves a draft Guide which will eventually have formal
approval.  Pat will take up the issue with Ann Kerr, head of the IEEE
Mass Storage Systems and Technology Technical committee, with the goal
of issuing a statement which vendors may use when dealing with these
requests.

Bob Riddle discussed the Andrew File System (AFS).  It maintains one
authoritative copy of a file and many cached copies.  Cached copies of
files are managed in 64 kilobyte chunks.  It permits multiple readers
of a file and only one reader/writer.  All readers get call-backs when
a reader/writer modifies the file.  It uses a variety of servers: a
Basic Overseer Server, a File Server, a Volume Server, a Volume
Location Server, a Protection Server, an Update Server, an
Authentication Server and a Cache Manager.

Sam Coleman, Peter Lawthers and others observed that the cacheing
function implemented in AFS is missing from the MSS Model.  Pat Savage
maintained that when viewed properly, it is not missing, rather it is
the case that the cache manager is the Model's client, under AFS.  It
was generally agreed that the Model did not encompass the call-back
feature of AFS.  Andy Hanushevsky talked about a system which features
an "intermediate server" which speaks NFS to its clients, uses AFS as a
middle layer, and uses the Model as a lower layer.

The meeting was recessed at 4:40 PM, July 25, and again brought to
order at 8:00 AM, July 26.

Steve Louis presented some concerns about the Model's treatment of what
it calls the "Site Manager." Discussion focused on the function and
method of the Site Manager in a heavily distributed system.  A
conclusion was that the Site Manager should henceforth be renamed the
"Storage System Manager" because even when a mass storage system is
heavily distributed, the resource being managed is storage, not any
other element of the distributed system, such as the network.

Loellyn Cassell presented a set of concerns about the Model's view of
Physical Volume Repositories (PVR).  She found it deficient in several
areas: inability to query the PVRReadQueue, the assumption that a
Volume Identifier is some label attached to the volume, missing
create/destroy functions for volumes, and lack of an insert function.
She also found it to be to vague, and suggested that the human tape
operator interface should be separated from the PVR.  She believed that
each mounting method (robots 1 ... n, and humans) should have its own
PVR.

Her presentation inspired a spirited discussion which converged upon a
proposal made by Dick Wrenn.  He suggested that the PVR model should be
broken into two layers.  The top layer, still called the PVR, should
define a name space for volumes, manage ownership, permissions and
characteristics (such as media type).  The lower layer should consist
of one or more instances of Robot Drivers.  Each Robot Driver would
control one mounting method (A's 3480 robot, B's 3480 robot, an optical
robot, a set of tape operator messages and commands, etc.).  Multiple
PVR's would be permitted and would define as many removable volume name
spaces.  An additional suggestion was made that the Model should
indicate that PVRs are not required for non-removable media.

Dennis Reed presented some suggested PVR Model refinements derived from
STK's experience implementing what the previous discussion identified
as a Robot Driver.  They included the desirability of

    -	specifying the PVR in a general enough way that it could support
	extra-model clients (as a tape library management system),

    -	letting the Robot Driver choose which of a pool of scratch
	volumes to choose (for mount optimization),

    -	supporting multiple "pools,"

    -	supporting multi-drive/multi-volume reservation (for deadlock
	avoidance), and

    -	maintaining ownership and access control for physical volumes.

Sam Coleman initiated a discussion of future meetings.  The next
meeting was set for September 17 and 18, at NCAR again.  Starting time
for the first day was set at 8:30 AM, and the ending time for the
second day was set at 2:00 PM.  Additional meetings will be held at
approximately 2-month intervals thereafter.  The November meeting will
be held at a time near to Supercomputer '90 (which will be in New York
City, from November 12 through November 16).

Action Items were assigned for the next meeting:

    1.	Janet Anglin will give a presentation on IBM's view of the Model.

    2.	Loellyn Cassell, Dennis Reed, Dave Tweten will present a
	straw man revised PVR model, after consultation with Richard
	Wrenn.

    3.	Andrew Hanushevsky and Bob Riddle will provide an explanation of
	the needs placed upon the Model by the Andrew File System.

    4.	Richard Wrenn will discuss as much as he is permitted of DEC's
	storage reference model.

Pat Savage explained how he would like to see the interface between the
Storage Server and the Bitfile Server changed.  He said that the
current model features a long hand-off path between the Bitfile Client
and the Storage Server's half of the Bitfile Mover.  He wanted the
Bitfile Server only to generate authentication tokens for use by the
Bitfile Client in sending requests to the Storage Server to read or
write the bitfile.  He wanted both ends of the Bitfile Mover to be the
trusted agents of the Storage Server.

Richard Wrenn asked how Pat's repartitioning of the Model would handle
a bitfile which is migrated between the issue of an authentication
token and the attempt to read or write the bitfile.  There were other
objections to the shift of complexity from the Bitfile Server to the
Storage Server.

The meeting was adjourned at 4:00 PM, Thursday, July 26.

Thanks to Bruce Griffing for maintaining the minutes, in my absence, to
the first break in Wednesday's session.

Dave Tweten
Secretary