Re: type of striping

From: Howard J. Rogers <hjr_at_dizwell.com>
Date: Thu, 09 Dec 2004 06:48:05 +1100
Message-ID: <41b75a73$0$12876$afc38c87@news.optusnet.com.au>

Richard Foote wrote:
> "Noons" <wizofoz2k_at_yahoo.com.au> wrote in message
> news:1102498989.134881.223450_at_f14g2000cwb.googlegroups.com...
>

>>Howard J. Rogers wrote:
>>
>>
>>>Because if you choose your uniform size poorly (ie, too small), then
>>
>>the
>>
>>>number of extents a segment acquires will shoot up. And the ASSM
>>
>>bitmaps
>>
>>>  get more burdensome the more extents they have to manage/deal with.
>>
>>Yes, but that would be the case as well with LMT without ASSM,
>>wouldn't it?  I mean, it uses a bitmap as well to allocate
>>space in uniform extents.  Too small an extent, too large a bitmap,
>>*potential* problem.  Does it get compounded with ASSM?
>>

>
>
> Hi Nuno,
>
> It's been a while since I've looked at all this in any detail and I don't
> have time to research my research (I've a new Pink Floyd DVD to investigate
> :) but to briefly answer your questions.
>
> The ASSM bitmaps reference *blocks* not extents and as such bitmaps can be
> stored in one block and yet reference/span a number of smaller extents or
> can be stored in several blocks in order to reference the blocks in one
> larger extent. However, the ratio of bitmap blocks to referenced blocks can
> be higher for smaller extents sizes although there are quite a number of ifs
> and buts with it all (for example results varied between uniform and
> autoallocate).
>
> With regard to LMT without ASSM and the *extent* management bitmaps of LMT,
> note that Oracle allocates a number of blocks (64Kish) in the first datafile
> of the tablespace regardless so providing these blocks are sufficient to map
> all extents in the tablespace, then extent size doesn't really matter. If
> however you have a massive tablespace with a smaller extent size, then
> Oracle may need to allocate additional bitmap blocks in the tablespace,
> where it's again dubious whether such an overhead is significant to really
> matter as well.

"dubious"... it's not a very scientific word, is it?! "28% of my buffer cache consumed by ASSM BMBs"... that's a bit more like it (I think it was Jonathan that posted that once... I hope he will clarify), and would qualify as significant in anyone's book, surely.

Regards
HJR Received on Wed Dec 08 2004 - 13:48:05 CST