From: "Jonathan Lewis" Subject: Re: Oracle Parallel Server and 24x7 availability Date: 2000/06/06 Message-ID: <960305947.21689.0.nnrp-09.9e984b29@news.demon.co.uk>#1/1 X-NNTP-Posting-Host: jlcomp.demon.co.uk:158.152.75.41 References: <_C8_4.231578$Tn4.2063175@news1.rdc2.pa.home.com> X-Trace: news.demon.co.uk 960305947 nnrp-09:21689 NO-IDENT jlcomp.demon.co.uk:158.152.75.41 X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3 Newsgroups: comp.databases.oracle.server X-Complaints-To: abuse@demon.net You are thinking of 'cache fusion'. This is a feature to reduce pinging due to read/write contention, not write/write. Oracle has implemented a 'consistent read server', so that if node A want to read a block that has recently been updated (and not written) by node B then node B will create a consistent read image of that block and pass it down the interconnect rather than writing the block, and the associated rollback segment header and rollback blocks that node A would need to create a CR image for itself. It node A want to write to the block, node B still has to ping it down. -- Jonathan Lewis Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk Bob Fazio wrote in message <_C8_4.231578$Tn4.2063175@news1.rdc2.pa.home.com>... >This will be short and sweet. > >I use it. I like it. I would recommend it. Raw devices aren't that hard, >infact I prefer them. It's not what is advertised, but it's still worth it. >In oltp enviroments you will want to be cautious, although I read that 8.1.6 >does have mode that eliminates the problem with pinging that can occur in >oltp enviroments. >