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: Oracle Parallel Server implementation questions

Re: Oracle Parallel Server implementation questions

From: koert54 <k_at_k.com>
Date: Mon, 06 Aug 2001 22:37:55 GMT
Message-ID: <7%Eb7.16983$lB.3337055@afrodite.telenet-ops.be>

"Sybrand Bakker" <postbus_at_sybrandb.demon.nl> wrote in message news:997127948.18371.0.pluto.d4ee154e_at_news.demon.nl...
> Answers embedded
>
> Hth,
>
> Sybrand Bakker, Senior Oracle DBA
>
> "tony sheng" <tonys_at_csee.umbc.edu> wrote in message
> news:aa15d4ec.0108061123.6bda30b2_at_posting.google.com...
> > Hi everyone,
> >
> > I have recently been asked to implement Oracle Parallel Server.
> > I am an ex-Sybase DBA and little knowledge about Oracle,
> > but figured this would certainly be a fun experience...
> >
> > Anyway, I have a few questions about implementation:
> >
> > 1. Tips on administration?
> > is this much more difficult than Oracle without the
> > parallel option?
>
>
> No
>
> You will need to assign public rollback segments though,
> make sure each instance still has some private rollback segments.
> If you are forced to perform recovery, remember to recover *both*
 instances.
>

You don't need to - but you can - use public rollback segments. If a node dies the surviving node will perform the instance recovery ... all other recovery is identical
to a single instance recovery - just start one instance and recover the DB ...

Backups ! Either use dd - but the easy way is to use RMAN so you don't have to worry about the raw devices ...

Oh ... and read up on PCM locks before firing up the DB (unless you use 9i)

> >
> > 2. disk layout and user activity?
> > i take it being careful how you layout the disks,
> > objects, etc. is important.
> > is oracle pretty good about managing multiple
> > instances trying to write to the same disks?
> > can you monitor how much thrashing is going on to the
> > same disk?
> >
>

Depending on OPS version (7, 8.0, 8i) watch out for pinging ! All the rest is pretty basic ...

>
> The operative word here is many OPS implementation on Unixes force you to
> use RAW filesystems (ie non-cooked)
> This doesn't apply however, AFAIK, to NT and OpenVMS

Last time I used OPS on NT a needed wolfpack and partitions instead of raw devices - but OPS on NT is nothing more
than a fun toy...
Compaq now has a clustered filesystem - no more raw devices in True64 clusters ! Even the root
filesystem is shared !

> > 3. any other best practices or big caveats?
> >

Yeah - pray and hope your application scales ! ...

> > thanks in advance,
> > tony
>
>
Received on Mon Aug 06 2001 - 17:37:55 CDT

Original text of this message

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