Re: 30 instances on one host

On Tue, 18 Jun 2002 04:01:17 GMT, Sean M <smckeown_at_earthlink.net> wrote:
>Alan wrote:
>>
>> For example, OFA asks the DBA to place software on a mount point
>> separate from the data mount points. The faulty logic behind this is
>> that it will separate I/O and thus boost performance. The truth is
>> that the software mount point will generate relatively little I/O; the
>> software mount point would be quiet and the data mount points would be
>> busy. So, a DBA following this advice would not optimize the I/O
>> potential of the host.
>>
>> The OFA advice on mount points also holds little water when viewed
>> from the perspective that on most decent sized UNIX boxes, I/O
>> spreading is now done inside the storage array. The days of having a
>> mount point logically connected to a disk partition's address are
>> gone. OFA was written for systems built before the proliferation of
>> disk volume management software like Veritas.
>
>This is not accurate. OFA is designed to address managability of Oracle
>isntallations by providing guidelines for naming conventions and the
>like. It is not designed to address performance. The reason to
>separate Oracle software is not for performance, but for managability.

It was indeed explicitly about performance, when the idea was to split I/O among physical disk devices. Eventually the reasons were watered down as it became cookbooked to /u001.../unnn, and completely stupid as modern storage hardware became prevalent. I just grabbed the old 7.0 Ault book p. 12, summarizing 2 of the 3 rules of the OFA process have performance implications; includes separating groups of segments that will contend for disk resources (e. g. data and indexes); and controlfile, redo log/group, and data that would have contention.

>
>> One bad naming convention in OFA is this:
>>
>> If you install an OFA-compliant Oracle Server, the Oracle home
>> directory is
>> $ORACLE_BASE/product/release_number.
>>
>> The above convention restricts the DBA from running 2 instances (or
>> 30!) of the same release on one host.
>
>Only if you assume each instance must have its own code tree. But you
>can certainly run more than one instance per code tree.
>
>That said, the reasons for running so many Oracle instances on the same
>host are few and far between. I can only think of one organization in
>our company that can justify such a configuration: training, where each
>member of the class needs his or her own database for the exercises.

I can additionally imagine a large org having a huge machine with a bunch of different depts with their own instances, for various reasonable reasons. (Not to mention unreasonable reasons, but I won't go there). I agree that it would be pretty rare. Although, doesn't IBM run machines with thousands of individual linux (linices?)...

jg

-- 
These opinions are my own. 
http://www.garry.to                                       Oracle and unix guy.
mailto:joel-garry_at_nospam.cox.net                       Remove nospam to reply.