Tuesday, November 28, 2006

Why I still use pfiles, revisited...

Ivan Kartik has an informative guide on SPFILES and their use at http://ivan.kartik.sk/index.php?show_article=22. I hope he posts a followup explaining what advantages they have over PFILEs and some real-world scenarios. Perhaps he can persuade me to change my mind...

5 comments:

Noons said...

it's gonna be darn hard to find a convincing argument to use them instead of pfiles, that's for sure...

other than the old ho-hum "because we say so".

Jeff Hunter said...

hey, I'm open to suggestion. I figure Oracle created them for some reason, I just haven't found it yet.

Dave said...

well one advantage is that you can start your database up from anywhere, with pfiles you have to be local to the machine.

rman can back them up

but no they dont offer much more. but conversely pfiles dont have any advantages over spfiles, so why not go with oracle and use them?

Noons said...

Actually: if you store the pfiles in a shared directory you can open databases from anywhere as well. Which you should be doing if starting from anywhere is the requirement.

On the other hand, one could argue that starting a database is best acomplished from the node where it is supposed to be executing. Given that it lets you verify the logs to ensure it actually started AND stayed started.

So, there goes another argument in the "for" case.

As for pfiles not having any advantages over spfiles: did you miss the points raised by Jeff? I think he explained rather well why pfiles are a better option.

"going with Oracle" is surprisingly the last thing anyone with more than a coupla years of using their software would want to do, believe me.

Joel Garry said...

I'm guessing the unspoken rationale has something to do with the interactions between control software and multiple instances. Imagine a very complex RAC configuration controlled by a GUI and some of the init parameters automagically tuned. There might be an argument there for not allowing simian involvement.

Personally, I'm totally against non-text configuration, whether for an OS or a db. But I can see there may be some level of complexity where one would want to make it difficult to modify, or maybe the real requirement should be strongly auditable. That level occurs when simple changes by the unknowing break the system in a manner that can't be repaired in a timely manner.

The fact that spfiles don't reduce to pfiles in a simple configuration is just a goof by Oracle. I've speculated this fascination with geeky registries has something to do with memes: VMS did it that way, so MS did it that way, so someone in Oracle decided to do it that way. Perhaps having config files all over the place in apache and such demonstrates that one could miss the benefits of centralized configuration, too.