home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!sun-barr!ames!agate!triplerock.CS.Berkeley.EDU!mao
- From: mao@triplerock.CS.Berkeley.EDU (Mike Olson)
- Newsgroups: comp.databases.theory
- Subject: Re: What is the _Halloween Problem_ ?
- Date: 29 Aug 1992 17:46:13 GMT
- Organization: University of California at Berkeley
- Lines: 27
- Message-ID: <17od55INNih8@agate.berkeley.edu>
- References: <1992Aug28.161050.668@news2.cis.umn.edu> <17loktINN7mi@agate.berkeley.edu> <BtpLzu.Mzx@world.std.com>
- NNTP-Posting-Host: triplerock.cs.berkeley.edu
-
- In <BtpLzu.Mzx@world.std.com>, edwards@world.std.com (Jonathan Edwards)
- writes:
-
- > In article <17loktINN7mi@agate.berkeley.edu> mao@triplerock.CS.Berkeley.EDU (Mike Olson) writes:
- > >transaction semantics force us never to see our own updates, so this is
- > >a violation of transaction semantics.
- >
- > Why's that? Seem perfectly reasonable to me that I should see my own updates.
- > How does this follow from the conventional ACID definition of transactions?
-
- if you want a high level of consistency in transaction processing, you must
- supply repeatable reads during a transaction. this means that inside a
- single command in a single transaction, you cannot be allowed to see your
- own updates.
-
- lower levels of consistency don't have this requirement, and (as a result)
- don't necessarily guarantee serializability of concurrent transactions.
- this violates isolatability (the 'i' in acid).
-
- jim gray wrote a paper in the early- or mid-80's defining four consistency
- levels, and giving the characteristics of each. the ref is at work (and
- i'm not); email me if you want it, and i'll dig it up.
-
- mike olson
- project sequoia 2000
- uc berkeley
- mao@cs.berkeley.edu
-