home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.databases
- Path: sparky!uunet!das.wang.com!wang!news
- From: MARK.CHILENSKAS@OFFICE.WANG.COM ("Mark Chilenskas")
- Subject: Re: Transaction Roll-Forward...
- Reply-To: Chilenskas@Office.wang.com
- Organization: PACE Database Group
- Date: 21 Jul 92 09:02:26 EST
- Message-ID: <199207211303.AA25318@tuna.wang.com>
- Sender: news@wang.com
- Lines: 32
-
- CCA in Cambridge Massachusetts were doing research on a similar line about 10
- years ago. Basically, the contract was figuring out how one could merge
- transactions after a network partition rather than blocking some or all of
- the sites on the network. Two approaches were examined, one looking at the
- semantics of individual transactions and one looking at the semantics of
- changed data items. Logs were merged individually at each site, and
- transactions run where discrepencies were found.
-
- In general, you cannot simply roll forward the journal from separate sites
- and expect a useful database. The problem is that the cumulative effect of
- transactions will violate integrity constraints that would have rejected
- some of them had they all been run serialized at a central site. You ship
- more product than you have inventory, end up with negative bank balances,
- etc. Within each application domain there seem to be generally accepted
- practices for dealing with such constraint violations, like charging the
- extra $20 for the overdraft, but they are application domain specific.
-
- I know of no general commercial application of this type of technique.
- However, a New York and a London bank (sorry, i don't remember which ones)
- were interested in the work and may have incorporated some of the ideas in
- their automated teller systems. Also, the WWMCCS inventory application used
- the data item semantics. I think logitics writing notes on the blackboard
- so capabilities understood that some bases were currently over booked was one
- type of compensating transaction used there.
-
- For more information, i would look for papers by Hector Garcia Molina and
- / or Susan Davidson for the data oriented approach, Barbara Blaustein and /or
- Dale Skeen for the transaction semantics approach, and Charles Kauffman for
- retrospectives on the experiment. They would tend to be 1982 to 1987.
-
- R Mark Chilenskas
- Chilenskas@Office.wang.com
-