gupta@fornax.UUCP (Ranabir Gupta) (02/20/91)
I'm looking at a planning/design application where lots of 'what-if' (hypothetical) design-related planning is involved using information about 1000+ objects. What complicates matters is that other planners will also be looking at the data (and wanting to modify it) and the world which the data models will also be changing and wanting to update the data to reflect the changes. Many updates may be considered before some are committed to the database by the planner. In a sense, each transaction will be a long one (upto hours) instead of the quick commercial transactions. Otherwise, the requirements of maintaining integrity of the data during the transaction are (almost) the same as in the usual case. Obviously, all the data which may possibly be updated in the course of the transaction should not be locked, for fear of hobbling all the other planners. The alternative that seems best is to develop some sort of warning mechanism to simultaneous accessors of the same information for read and update (even for read, because subsequent plans may depend on a planner assuming that some data which (s)he is looking at won't change during her/his deliberations) and some conflict resolution strategy. Question is: Are there any systems around which target this sort of application ? Are there any research papers that do ? Partial addressing of this question ? I am not bound (at present) to any particular environment (object- oriented/relational etc.) but would 'prefer' one which leans towards established technologies. Gurus, help! if people email me at 'gupta@cs.sfu.ca', I'll summarize for the net, if there's sufficient interest. Of course, any response could just as well be posted. Thanks in anticipation.
davem@mtunion.SanDiego.NCR.COM (Dave Miller) (02/20/91)
In article <2151@fornax.UUCP> gupta@fornax.UUCP (Ranabir Gupta) writes: >I'm looking at a planning/design application where lots of >'what-if' (hypothetical) design-related planning is involved >using information about 1000+ objects. What complicates matters >is that other planners will also be looking at the data (and >wanting to modify it) and the world which the data models will >also be changing and wanting to update the data to reflect >the changes. > [more description deleted] > >Question is: >Are there any systems around which target this sort of application ? >Are there any research papers that do ? >Partial addressing of this question ? > MCC - Microelectronics and Computer Technology Corporation - is currently researching new transaction techniques that relax the ACID properties as part of the Carnot project. NCR is a sponser of the project. I don't know what is / is not confidential so I had better shut up now. Dave Miller
cwk@ORCRIST.GANDALF.CS.CMU.EDU (Charles Krueger) (02/23/91)
> Are there any systems around which target this sort of application ? > Are there any research papers that do ? > Partial addressing of this question ? We designed and implemented a long transaction facility as part of the object database in the Gandalf Software Environment Generator. It seems to satisfy all of the requirements in your post. I've have a tech report that describes the issues and implementation. The tech report is written in general terms, so it's applicable to most object-oriented database models, not just software databases. Here's the citation: TechReport( Title="Persistent Long-term Transactions for Software Development", Author="Charles W. Krueger", Institution="School of Computer Science, Carnegie Mellon University", Address="Pittsburgh, PA. 15213" Number="CMU-CS-90-188", Month="December", Year="1990") If you are interested in getting a copy of this tech report, you can send me e-mail. (If I get too many requests, though, our document manager might make me charge you a couple of dollars!) Also, Sun's Network Software Environment (NSE) has the flavor of long transactions. It uses optimistic concurrency and a merging tool to assist software developers when there is a concurrency clash. It is oriented towards fairly large-grained (file-sized) software "objects". Here are a couple of citations: Book( Author="William Courington, Jonathan Feiber, and Masahiro Honda", Title="NSE Highlights", BookTitle="The SunTechnology Papers", Chapter="3", Publisher="Springer-Verlag", Year="1990") -------- Proceedings( Author="Evan W. Adams, Masahiro Honda, and Terrance C. Miller", Title="Object Management in a CASE Environment", Organization="ACM SIGSOFT", BookTitle="Proceedings of the 11th International Conference on Software Engineering", Address="Pittsburgh, PA.", Month="May", Year="1989", Pages="154-163")
gupta@fornax.UUCP (Ranabir Gupta) (02/28/91)
Thanks to all those who responded to my query about products/research which
support long (upto days) transactions on shared objects by many users in a
design/planning environment.
I'd offered to summarize on the net - here's the summary. One person sent
email asking me to mail him the summary. Sorry, I deleted your email by
mistake. I hope you're reading this ...
------------------------------------------------------------------------
The problem:
1) Many 'objects' shared by many planners/designers.
2) Designers want to commit updates to sets of these objects en masse, much
like a transaction.
3) Catch: Designers would like to examine these objects for days before
committing their transactions.
4) This leads to following hassles:
a) can't lock objects accessed by one user for fear of crippling all the
others' work.
b) user A must have some way of knowing which other users are accessing
or planning to modify the objects (s)he's looking at, and must be
notified of (impending) changes to those objects, so that days worth
of planning does not get washed away at commit time when (s)he discovers
that some object has been changed 'on the sly' while (s)he wasn't looking
-------------------------------------------------------------------------------
Responses:
(In general, this is not a properly explored area. Most attempts seem to be
stabs at part of the problem)
Research (Academia):
1) Matthias Jarke and group are working on ConceptBase at University of
Passau. This system contains some of the features involved. It can
support design groups spread over continents (over Internet). They
are writing a "Conceptbase V3.0 manual". Contacts:
jarke@unipas.fmi.uni-passau.de
jeusfeld@andorfer.fmi.uni-passau.de
2) Using ConceptBase to implement cooperative software development
(GroupWare) is an activity under John Mylopoulos at University of
Toronto. Possible contact:
rose@telos.ai.toronto.edu
There's apparently some more interest in the subject there.
3) At CMU, a long transaction facility has been implemented as part of the
Gandalf Software Environment Generator, which seems to satisfy most
of the requirements. ref:
Charles Krueger, Persistent Long-Term Transactions for Software Development,
Tech Rep. CMU-CS-90-188, School of Computer Science, Carnegie Mellon
University, Pittsburgh, PA 15213, Dec, 1990
4) Other refs:
a) Won Kim, HF Korth, A Model for CAD Transactions, Proc. of VLDB 1985,
Stockholm, 1985
b) Henry F Korth, W Kim, F Bancilhon, On Long Duration CAD Transactions,
Information SCiences, v46, 1988 pp73-107
c) Henry F Korth, Long Duration Transactions in Software Design Projects,
Proc. 6th Int'l Conf. on Data Engineering, Feb 1990, Los Ang. CA
d) Patrick E O'Neil, The Escrow Transactional Method, ACM Transaction on
Database Systems, V11, no4, 1986, pp405-430
Commercial Systems:
1) William J. Doucette
Object Design Inc.
1450 E. AMerican Lane, Suite 1400
Schaumberg IL 60173
This company has a product which has an explicit mandate to support
distributed design and subcequent reconciliation of the resulting
designs.
2) Sun's Network Software Environment has a flavour of long transactions. It
is oriented towards file-sized software 'objects'. Refs:
a) William Courington, J Feiber, M Honda, NSE Highlights, Chapter 3, The
SunTechnology Papers, Springer-Verlag, 1990
b) Evan W Adams, M Honda, TC Miller, Object Management in a CASE Environment
Proc. ACM-SIGSOFT 11th Int'l. Conf. on Software Engineering, Pittsburgh,
PA May 1989
c) (Mysterious One) MCC and NCR may be doing something in this area.
Thanks to:
Manfred Jeusfeld
Chris Arnold
Vinay Chaudhri
Charles Krueger
Dave Miller
-----------------------------------------------------------------------------
Disclaimer: I can't imagine why someone would like to nail me or anyone else
for any of this, but hey! this is america, and so I disclaim all
knowledge of the constructive/destructive/soporific properties of
any of the products/papers mentioned herein, so help me God.
Ciao,
thanks again, ranabir