home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!news.gtech.com!noc.near.net!transfer.stratus.com!redondo.sw.stratus.com!dswartz
- From: dswartz@redondo.sw.stratus.com (Dan Swartzendruber)
- Newsgroups: comp.sys.stratus
- Subject: Re: loss of end-of-file mark problem
- Date: 9 Jan 1993 15:59:36 GMT
- Organization: Stratus Computer, Software Engineering
- Lines: 37
- Message-ID: <1imsp8INNchm@transfer.stratus.com>
- References: <1993Jan7.131833.15374@porthos.cc.bellcore.com> <k403wB1w164w@cellar.org>
- NNTP-Posting-Host: redondo.sw.stratus.com
- Keywords: tps
-
- In article <k403wB1w164w@cellar.org> ibycus@cellar.org (Walt Mankowski) writes:
- >dante@decoy.uucp (25439-dante) writes:
-
- [description of ESD deleted]
-
- The only problem with ESD is that one can't always count on it. ESD will
- only be performed if:
-
- a) there were no hardware problems.
-
- b) there was no wired heap corruption.
-
- c) the disk subsystem isn't hosed.
-
- d) the disk cache subsystem isn't hosed.
-
- I'm not saying this to denigrate ESD. It is a very useful bit of
- bulletproofing, but that is all it is. If a customer absolutely
- can't afford to take an index-corruption hit, he needs either TPF
- or LPF. Which option they elect to use depends on a number of
- performance and reliability issues. ESD is especially not guaranteed
- to prevent the type of index block corruption described, since it
- can only flush blocks the higher-level subsystems have written.
- It is unlikely (but possible) that the index manager could be in the
- middle of a multi-block operation (splitting index blocks?) and has
- written some subset of these blocks to the disk cache.
-
- >Walt
- >ibycus@cellar.org
- >ibycus%cellar@tredysvr.tredydev.unisys.com
-
-
- --
-
- #include <std_disclaimer.h>
-
- Dan S.
-