Joel Garry wrote:
> It doesn't. It is smart enough to parallelize and get part of the log
> from one member and part from the other!
>
>>From an old O7 faq:
>
> "Having multiple members to a logfile group is also advisable. If
> there
> are multiple members, all members in a group are used to perform the
> archive process. Assuming that there are three members, the first
> chunk
> is read from one of the members and is being written to the archive
> while
> reading a second chunk from second member, then a third chunk from
> third
> member, then back to first and the process continues in round robin
> fashion.
> If a corruption is found in one of the members, it is validated (read
> again)
> and if the corruption still exists, reading from that member is stopped
> and the rest are used.
>
> This is one of the big benefits of using multiple members instead of
> mirroring at o/s level. Oracle knows about the multiple members so it
> can
> optimize on archiving, but it does not know about the mirrors. One
> other big
> benefit with using multiple members is that a separate write is issued
> for
> each member, so the odds of having a totally corrupted redo log is
> diminished (corruption written to one mirror will usually propagate
> over to
> all other copies).
> "
>
> I have personally seen it figure out and overcome fairly severe
> corruption problems in recent versions, with no noticeable effect to
> users. But definitely check your alert log every day, there would have
> been a big effect one time if I hadn't noticed the messages about the
> failing hardware in time.
>
> jg
> --
> @home.com is bogus.
> ohnohackers!
> http://www.signonsandiego.com/uniontrib/20050418/news_lz1b18hacker.html
>
Joel - I cannot see round-robin and parallel together; one is serial,
the other not (obviously :) ).
Your statement would mean the archive logger would be faster when
more members are being used - I don't think that is true.
Could you clarify, please?
--
Regards,
Frank van Bortel
Received on Wed Apr 20 2005 - 08:12:01 CDT
