Re: Parallel processing advantages

in general, performance should benefit from parallel query (which is different from parallel server). the key factors you need to consider are:

1. frequency of queries
2. number and type of queries
3. range of the size of result sets (something like a bell curve...what % have how many rows returned)
4. speed of the disk, controller, and i/o bus and latency potential between reads

if one recognizes that there may be latency between requests or even during a single request, then multiple queries can take advantage of the latency. also, if queries are returning similar result sets, the data may be cached in the i/o card memory (if it has a cache). sometimes dbas tend to overlook the fack that "contiguous blocks" from oracle's perspective may be fragmented physical blocks on a disk, as oracle works above the physical disk layer (unless raw i/o is being used, which most of the time is not). so the disk heads are most likely bouncing around no matter what you do with logical space.

steve long
"The world's best technology strategist." (TM)

"Scott Watson" <swatson_at_datachest.com> wrote in message news:pojB6.642289$JT5.17477940_at_news20.bellglobal.com...
> I would like to know if there is any advantage to using parallel
 processing
> in the following situation. Now before you tell me that it would be
 better
> to stripe the table across disks and the like I realize that, however, I
> wanted to know if there was any advantage without changing the table
> whatsoever.
>
> Situation.
> 1 table 10,000,000 rows.
> Stored in one tablespace with all datafiles on the same disk.
> Access is read only
>
> Is there any performace increase when scanning a table that is on the same
> disk in parallel or is it better to scan it serially.
>
> My thinking would be as follows. If there was only one user accessing the
> table, and the table was the only object stored on the disk in question
 then
> parallel processing would not be faster. However, assuming that this is
 not
> the only object on disk and that multiple user may be using the table at
 any
> given time then wouldn't PQ be better as n slaves would be able to read
> n*mulit block read count bytes on each read. Therefore getting through
 the
> table quicker all the while services other request from the same disk.
>
> Thanks for your input.
> Scott Watson.
>
>
>
Received on Thu Apr 12 2001 - 16:24:28 CDT