Re: Problems that can ocuur while upgrading from 9i 1o 10g

On Sep 18, 3:58 am, Jenny <jjohni..._at_gmail.com> wrote:
> Our company is planning to uprage from oracle 9i to 10g.
> I want to know what all problem i will face during upgradation..

See http://www.oracle.com/technology/products/bi/db/10g/pdf/twp_bidw_optimizer_10gr2_0208.pdf

I just completed one this weekend (plus a hardware change and app upgrade). Since the app upgrade part of this was very involved, I had the luxury of taking my time with the oracle side of it, and barely remember the difficulties, except for a few very recently.

So just off the top of my head:

Justify all init.ora changes and settings. My app is profligate with cursors, so the vendor recommends 8000. I had gone much lower when I went to 9 because of the advent of session_cached_cursors, but now had to go back up. Had problems with shared pool when I ramped up to full usage, so had to increase it (app uses bind variables). Was a problem, because I had set sga_max_size to allow similar sga size to my previous system (D'Oh!), and until I could bounce had to reduce buffer cache so I could increase the shared pool. Log buffer sizing is different, Oracle uses memory granules (read the docs), so you wind up with many megs, it can be quite different than what you set. Be especially vigilant if you have any underscore parameters set. Be sure you understand the issues about async I/O for your configuration, it varies a lot. Be sure you understand sort area parameters and the pga_aggregate thingie.

The EM advisors seem pretty good, within limits. One shortcoming is some things seem to just want more and more. I think the performance pack is worth the cost, assuming you don't use it to avoid learning how things really work. Google for ASH simulation if you are standard edition.

I've been told optimizer_index_caching may be more trouble than it is worth, so I removed that from init.ora. I've found index_optimizer_cost_adj appears to be useful.

RMAN is getting pretty sophisticated, but I still haven't figured out how to get it to do exactly what I want entirely within Oracle, but I haven't finished setting up the repository for it yet either. I'm quite impressed with it's compression and speed, and am mostly happy with the EM interface, though obviously if I had written it I would have done some things different. Then again, why think I am smarter than Oracle developers? :-) I'm still seeing strange unexplained usage spikes, and EM tells me the OS is swapping when I don't see it...

I did have problems with patching to 10.2.0.4, most stemming from me not being paranoid enough (!) and checking to see dbconsole shut down when it said it did. I'm still having some issues with subsequent instances starting up dbconsoles, haven't had time to figure it out. Be sure and understand all parts of the technology stack, especially if you haven't had much experience with apache or oracleAS type things.

So far the automatic jobs appear to be doing the right thing, but you might want to google around for the interminable discussions about things like how often to calculate statistics. The right way is sitedependent.   My own view on that has turned 180.

It made sense to use exp/imp due to the app. One of the advisors told me which objects had row chaining and wasted space due to the subsequent app upgrade, which I found very helpful, don't know when I would have gotten to it if I had to do it manually.

"Private strand flush not complete" is generally ignorable.

And I second everyone about reading the upgrade manual, as well as the new features docs.

jg

--
@home.com is bogus.
International talk like a pirate day, Sept. 19.  RRRRRRRR, man!