Plain simple.
Large trans => large rollback segment size and wise versa.
Initial must the be 21 and optimal the size of (the universal)
42 extents!
/Svend Jensen
Howard J. Rogers wrote:
> Funny, I don't disagree with any of this. The conclusion at the end is
> exactly what I'd say if I could be bothered to type it all. I guess my
> 'take' on it is that it's a tricky thing to get right, and the costs
> therefore mount. The shorter version is thus that 'you are unlikely to get
> it right, so avoiding the costs is the better approach'.
>
> Maybe I'm just pessimistic.
>
> ;-)
> HJR
>
>
> "Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote in message
> news:angshg$dn$2$8300dec7_at_news.demon.co.uk...
>
>>It's not often that I feel the need to disagree
>>strongly with Howard, but I believe that setting
>>Optimal (properly) is likely to be beneficial.
>>
>>Richard Foote wrote a reply on the newsgroup
>>recently explaining why having the smallest
>>possible volume of rollback can be an aid to
>>performance by reducing the total DBWR load,
>>so I won't repeat all the details.
>>
>>The concern that 'optimal' causes a massive
>>dump to disc 'just when you don't want it to
>>happen' is potentially deceptive.
>>
>>When a rollback segment shrinks, the dirty blocks
>>in the discarded extents have to be written
>>to disc (to cater for worst case crashes). This
>>is true, and this is the 'massive performance hit'
>>that you have complained about in the past.
>>
>>However, Oracle will dump the oldest extents
>>to disc, and if the optimal has been set properly,
>>these are the very extents which are likely
>>to have been dribbled out to disc by the
>>continuous checkpoint / MTTR algorithm
>>that Oracle uses nowadays. After all, the
>>I/O benefit you get from right-sizing is that
>>you don't have excess extents that are constantly
>>dribbling to disc.
>>
>>
>>Clearly, if you set the optimal a little bit too small,
>>then you are likely to be adding and dumping extents
>>frequently, and the extra dumping you do is likely to
>>be of extents that would not otherwise be written -
>>and your argument has some merit.
>>
>>Otherwise it is the standard Oracle argument of
>>which feature to use when:
>> I can pay a small price thousands of times per hour
>>or
>> I can pay a large(r) price once per day
>>
>>
>>Small rollback segments are good - if possible.
>>If they are possible, then Optimal is good.
>>
>>The price may be unacceptable, in which case
>>a 'manual optimal' may be the better approach.
>>
>>
>>--
>>Regards
>>
>>Jonathan Lewis
>>http://www.jlcomp.demon.co.uk
>>
>>Next Seminar dates:
>>(see http://www.jlcomp.demon.co.uk/seminar.html )
>>
>>____USA__________November 7/9 (Detroit)
>>____USA__________November 19/21 (Dallas)
>>____England______November 12/14
>>
>>The Co-operative Oracle Users' FAQ
>>http://www.jlcomp.demon.co.uk/faq/ind_faq.html
>>
>>
>>
>>
>>
>>Howard J. Rogers wrote in message ...
>>
>>>
>>>Don't set optimal. Please. It's crap for performance, and 1Mb is in any
>>>
>>case
>>
>>>extremely small.
>>>
>>>
>>
>>
>
>
Received on Thu Oct 03 2002 - 16:08:38 CDT
