Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Fail Safe Strategy

Re: Fail Safe Strategy

From: M.Rapier <M.Rapier_at_shef.ac.uk>
Date: 1997/02/17
Message-ID: <5e9h7d$p19@bignews.shef.ac.uk>#1/1

In article <3308071F.71AA_at_icanect.net>, program_at_icanect.net says...

{snip}

>I have begun to think about doing what I can to crash proof the production
 system -- and
>in the process playing some "What If" games... for example -- what if NT
 locks up and
>eats the redo log files...
>
>I am thinking that some kind of database mirroring or tandem operation across
 sets of
>drives or servers might be the best way to go. Failing that, some kind of
 automatic
>replication. Of course, there is no need to re invent the wheel here. Im
 sure folks
>have already thought all this out, and have similar systems functioning. I'd
 like to
>hear from folks who have considered this kind of thing.
>
>What is best? What works? ("Enquiring minds want to know" <grin!>).

We use full disk mirroring (after all, disk is cheap), we were going to do three-way mirroring (but disk isn't _that_ cheap), which has the added bonus that you can detach one of the mirrors for backups & get the system up again in a few minutes even for a full dump.

The online redo logs are protected by the disk mirroring, but they are also duplexed within Oracle (to avoid those annoying inconsistencies that can sometimes creep in).

Otherwise it is all common sense stuff out of the manual, multiple control files, archive log mode, decent UPSes etc.

The only thing to keep any eye on is your archive redo log directory as when (not if) it fills up, your DB will hang. Ours gets archived out to tape on a nightly basis, though it is sized up for a few days worth of transactions. It only needs a few big batch jobs to run each night, and you'd be astonised how many archive logs you get.

Replication, standby machines etc depend on how much unplanned downtime you can get away with. 99.99% availability is possible, but with exponentially increasing cost. Our users don't mind half an hour out per night and the odd weekend for maintenance. Luckily our hardware has been pretty good, we've had one power supply blow and one disk failure, which the mirroring handled fine.

Cheers.
Martin.

-- 
Martin Rapier, Database Administrator
Corporate Information & Computing Services.
University of Sheffield
-----------------------
To Infinity and Beyond.
         
Received on Mon Feb 17 1997 - 00:00:00 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US