Re: Multiplexing redologs - is there still need for it?

"Dusan Bolek" <pagesflames_at_usa.net> wrote in message news:1e8276d6.0403130127.7cb73d18_at_posting.google.com...

> As I thought about it, I think the point is if there is a real
> possibility that Oracle (with HW assistance) can write one redolog
> right and other corrupted.

It appears the problem is still there and possible, as there have been a few rumours it can happen with the latest versions. Worries me, quite frankly...

> level) went wrong. Well, that's possible, but If you have DG in place,
> you should be protected against this (primary fails, but secondary got
> all transactions right).

All that is nice. But if one has a non-multiplexed redo and triplexed(?!) control files like it is so usual, there is still a considerable overhead taken. In some situations, Oracle's control files happily single-thread transactions. If a non-multiplexed redo log is configured for performance reasons (can't see any other reason) and there are still 3 control files in non-optimal placements, "useless config" is the term that comes to mind.

Given the consistent inconsistency (darn weekends...) of redo log writing in Oracle, I'd say at least two redo log copies and a max of two control files. Assuming of course you have the adequate number of spindles to spread the load of all that synchronous I/O for both performance and reliability. And a manager that will understand why with that expensive EMC sitting there you are STILL multiplexing via software...

> And for the last HJR's line, I think that sometimes is good to open
> even "banned discussion". :-)

He'll crucify us soon, don't worry! :)

-- 
Cheers
Nuno Souto
wizofoz2k_at_yahoo.com.au.nospam