SSSWG - September 1995 Minutes

	IEEE Storage System Standards Working Group Meeting 
	14th September, 1995 
	Mass Storage Symposium, Monterey, California. 
 
 
Attendance: 
	Betty Jo Armstead (Sterling Software) 
	Jack Cole (Army Research Laboratory) 
	Jeff Deutsch (IBM) 
	Michael Diesburg (FermiLab) 
	Marty Frary (Storage Technology Corporation) 
	Bruce K. Haddon (Storage Technology European 
		Operations) 
	Merritt Jones (MITRE Corporation) 
	Dennis Luck (Department of Defense) 
	John Merrill (NCAR) 
	Gary Mueller (Storage Technology Corporation) 
	Rich Ruef (IBM) 
	Jamie Shiers (CERN) 
	Dave Skinner (Fujitsu) 
	Tim Sullivan (Hewlett-Packard Laboratories) 
	Rich Wrenn (Digital Equipment Corporation, 
		Co. Springs) 
 
Chair: Dave Skinner. 
Recording Secretary: Bruce K. Haddon. 
 
General Session, Thursday, 14th September, 3.35 p.m. 
 
The meeting was declared to be a special meeting. The 
agenda for the meeting and action items from the July 
meeting will be carried forward to the November meeting, 
and will be re-evaluated on the light of this special 
meeting. 
 
The agenda of this meeting was agreed to be: to discuss 
the question of the continuing and future work of the 
Storage Systems Standards Working Group. 
 
Dave Skinner introduced this topic with a resume of work 
and actions that have been undertaken since the 
publication of the Approved Draft Reference Model (the 
"Open Storage Systems Interconnect Reference Model") in 
September, 1995. 
 
Merritt Jones, the Working Group Sponsor, invited all 
present to speak to their expectations of the work, known 
reasons for attendance or support of the work to date, 
and to comment upon future participation and 
contributions. 
 
Dennis Luck said that the objective of his organization 
was to be able to buy storage system components 
competitively. 
 
Marty Frary spoke to the observed matter of the reduction 
of vendor support, by pointing out that a true solution 
to this problem (as stated by Dennis) would reduce the 
margins of the established vendors. 
 
Jamie Shiers said that although his organization wanted a 
component solution three or four years ago, they probably 
do not now. He still wants an architecture against which 
to judge the solutions proffered by vendors, and the 
contribution to understanding issues like scalability 
that the Working Group brings. 
 
Dave Skinner pointed out that the Reference Model is 
heavily referenced in mass storage publications (such as 
the 21 papers that did so in the 1995 Mass Storage 
Symposium, which had just closed). This is in spite of 
the existing and known flaws in the text and the 
underlying model. 
 
Rich Wrenn wished to disagree slightly, by pointing out 
that some of those references came from HPSS-related 
papers. However, the HPSS people no longer participate 
heavily in the SSSWG. Those that are left in the SSSWG 
are in fact almost all representatives from vendors. 
 
In general discussion, several people offered that the 
cuts in travel budgets are a big problem, and Government 
has been more affected recently, although commercial 
budgets were cut some time ago. 
 
John Merrill said the his organization had gotten what it 
needed, in that it used the Model to architect their own 
system, but were not interested in the generation of the 
actual standards. In retrospect, he feels they might have 
continued to have supported a further refinement of the 
Model. 
 
Merritt Jones told an anecdote of where "compliance" to 
the Model was required. This in impossible, as the Model 
is a "guide", and not a standard. 
 
Betty Jo Armstead pointed out that her organization has 
problems because components do not exist, and thus she 
must wait till her primary supplier integrates the new 
technology as it appears. There is no chance that the 
pieces will "just plug in". 
 
Dennis Luck observed that networking has altered the 
world in the last few years, and must be taken into 
additional consideration. This was agreed by Jamie 
Shiers, saying that services are the key. Betty Jo 
agreed, saying that some services are not of interest to 
her, such as PVL. 
 
Merritt asked the question: "Is working toward producing 
API's (by the end of the year) the right thing to do?" 
Jamie asked whether this would get to the point where, 
say, one product from X would work with one product from 
Y. Rich Wrenn asked whether these products were from 
user's or vendors, or whether the question was whether 
user's could write applications that worked with 
different vendors. Jamie said he thought that the answer 
to both was "no", but thought that there was a further 
question, which had to do with how many pairs of products 
(e.g., PVL - PVR ) he would need. 
 
Everyone agreed that having the standards would be 
better, but would not be the compelling argument. The API 
at the PVL level is most of interest, but an API is not a 
great help at the PVR level. Mary Frary made reference to 
the support problem. Betty Jo pointed out that it is 
becoming increasingly impossible to replace ALL of a 
storage system at once anymore, and this makes "plug and 
play" type functionality more necessary. 
 
Jamie still wants and needs "interoperability" interfaces 
for his Computing Centre. 
 
Rich Wrenn recounted a story which illustrated that, in 
general, users do not want PVL's and PVR's, but "backup" 
products, "archiving" products, etc., in which these 
components are integrated. They do not recognize nor wish 
to be concerned with the interoperability issue, which 
they only discover when they want to add new components 
or new functions. Jamie points out that the SSSWG's role 
may be to educate. 
 
In answer to a question, Rich Wrenn estimated two (full) 
resource-months work to finish a PVR API. An estimate was 
made by the other "regulars" that PVL is probably not as 
well placed, and that the MVR API is more complete, but 
the MVR-MVR interface is missing. 
 
The idea of floating a "consortium" was lightly 
discussed. Some organizations would be willing to 
contribute to, and this gives a measure of the "value" in 
which the work is held. 
 
Bruce read his list of "Version 6" issues (attached as 
Appendix A). This list was long enough that the general 
reaction was to ask the question of "what would be the 
one thing best on which to focus and to spend the small 
resource available?" Jamie suggested the PVR API was what 
he wanted. Because of lack of agreement on this 
suggestion, Merritt asked the question: "What are we 
doing here?" This led to the observation that if nothing 
was going to happen for the next 9 months, the work may 
as well stop here. 
 
A mini-survey was made of "user" representatives in the 
room, and they unanimously agreed (with only slight 
indications of disagreement) that the order of appearance 
should be 
		PVR - PVL - MVR. 
This will be taken as input to a decision that will be 
made determining the future work of the SSSWG. This issue 
will be taken up at the November meeting. 
 
(For reference, the meeting schedule from the July 
minutes is included here: 
 
*	Dates for 1995 
 
	July 18-21: DEC, Seattle. (Host: Michael 
		Peterson) 
	Sept. 14-15 (Monterey, at the end of the MSS) 
	Nov. 7-10 (ACSC, Los Angeles, Anit Gafni) 
 
*	Dates for 1996 (partial list) 
 
	Jan 16-19, Tucson, Arizona (Host: Eric 
		Stouffler, IBM). 
	Mar 12-15 (tentative, Colorado) 
	June (MSS is in Annecy, France). (Note added 
		later: the 1996 MSS planned for 
		Annecy in June has been postponed. A 
		new date is not yet available.)) 
 
The meeting adjourned at 5.07 p.m. 
 
Appendix A 
========== 
 
	Version 6 Issues 
	(Bruce K. Haddon) 
 
"Object-oriented" formulation. 
Security and a security model for all components. 
Management model and environment model definitions (in 
	particular, the relationship of these two, 
	including "OID"'s). 
Role of "policy", means of implementation in each 
	component of the model, and the management of 
	policies. 
Event notification between all components, and with the 
	management system; rules for the propagation of 
	errors, error handling and automated suppression, 
	logging, including all interactions with humans 
	(system administrators). 
Transaction semantics (including overall management of 
	transactions, and the related issues of atomicity) 
A metadata model, including storage of metadata in 
	parallel to the standard storage model itself, and 
	the support for recovery of metadata for all 
	components. 
Reintegration of a "data object" level into the Reference 
	Model (formerly known as "bitfiles") 
Cartridge transportation services, otherwise known as 
	"information interchange" (implying probably an 
	inter-PVL architecture and a method of representing 
	metadata between PVL's). 
Data export and import via other media (including 
	networks (with same metadata problems).
Interoperability ("plug and play"). 
Integration of fixed media with the Reference Model. 
Scalability issues. 
Remaining PVR issues (notably stackers and the like). 
Networking issues (protocols, data representation, 
	distribution of model components). 
Hierarchies of MVR's (assumed in Version 5, but not 
	described). 
 
Plus the issues concerned with the document itself, 
including a greater consistency of its structure and 
editing, and an effort to simplify the language.