Re: Excessive Logical and Physical I/O

>
> I think what he is asking is, if there is 1M worth of blocks to be
> consistent in a 1G RBS, might Oracle have to look through the whole 1G
> to get that 1M. And I think the answer would be the rollback headers
> track where the info is by block, so it shouldn't be a problem.
> http://www.ixora.com.au/q+a/undo.htm might help clarify some. And as
> you said, extent allocation is the real problem, so I would add, size
> the extents so the transactions fit in them.
>

Yes, that is exaclty what I was getting at but you have described it more explicity.... I will take a look at your recommended link on Steve Adams' site.

I understand the problem of tuning for tuning's sake, but I just wanted to clarify the problem with very large rollbacks (1G).

Thanks to Howards comments I can see that the only drawback to v large rollbacks is that new blocks that must be acquired will not be in the buffer cache, thus causing more i/o.

Thanks to everyone for the feedback..

Matt Received on Thu Mar 25 2004 - 07:25:05 CST