home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.databases
- Path: sparky!uunet!cs.utexas.edu!sdd.hp.com!hpscdc!cupnews0.cup.hp.com!dhepner
- From: dhepner@cup.hp.com (Dan Hepner)
- Subject: Re: Hot Standby DBMS's
- Sender: news@cupnews0.cup.hp.com
- Message-ID: <BtCM7B.FyG@cup.hp.com>
- Date: Fri, 21 Aug 1992 19:30:47 GMT
- References: <BtB5us.MsD@world.std.com>
- Organization: Hewlett-Packard
- X-Newsreader: Tin 1.1scd1 PL4
- Lines: 26
-
- From: edwards@world.std.com (Jonathan Edwards)
-
- >>Since all locks need to be obtained on both systems before the
- >>transaction is eventually committed, lock race conditions (defective
- >>design) result in deadlocks. The system as a whole remains perfectly
- >>serialized.
-
- >Lock races are inherent functionally: what happens when two operators want
- >to update the same record.
-
- While a race in one sense, this is a benign race unlike the disruptive
- race described by:
-
- >My point is that the winner will not necessarily be the same on a
- >system recieving a copy of the input stream.
-
- If so, your transactions will deadlock, and will not complete. In the
- system described, all locks must be obtained on all systems before the
- transaction can complete. The locking behavior is that of two unrelated
- databases. This is not a claim that there is no problem here: the races
- must be prevented in order to do any work, a well known problem in
- transactional application design. But that inability to do work is
- the penalty of screwing up, rather than an out-of-sync copy of the
- database or some other manifestation of non-serialized behavior.
-
- Dan Hepner
-