Irfhan had given a Good explain there and i thought i will not again go back to bickering about this .. But nisars comments on Disk Head movement and he considering an Extent as a Physical and not a Logical Component has spurred me to write this. [ i dont want anybody to think otherwise on this ] 1. A Block is a physicall storage and it may consists of mutiple OS Blocks .. 2. A Block is never contigous. 3. Blocks within an Extent are never contigous So now let us see where we stand with disk head and placement. :) Rgds, Ganesh On Mon, 8 Nov 2004 13:42:01 -0800 (PST), Nisar Tareen <ntareen@xxxxxxxxx> wrote: > Ganesh, > > I did not recall of mentioning Parallelism what I rote to Irfan, My answer > was about multiple extents. > > Yes, I will stand on my ground that multiple extents (as I said extra > ordinary) will affect the performance for full table scan, as the disk head > had to reposition itself > > To make extra hits (I/O) for read and write/update, and extra block > addresses in SGA for the records read. > > My understanding my logic, Ganesh most of the time I write from the top of my > mind, After doing this work for so long, I do not go through the technical > manual, My idea is not to write technical manual for every email or query, > Email should give the major guide line on the issue and each one of us the > user should go the extra 9 yards according to our needs. > > Believe me I ran Tara byte database and I don't have to prove, I had done my > proving, Few extents are better that many. > > I am not an academic and I am running a shop and the time I have I enjoy > answering emails and some time for fun add my logic. But Oracle is in the > business of proving yes I can be wrong then you just have to say, Nisar here > you are wrong. I will and you can give the understanding on an issue, its > okay with me. > > Plus, extents are Physical blocks not logical. eg. 1 block of 10 giga bytes > or 10 blocks of 1 biga bytes. > > Have Fun. > > > Nisar Tareen > > > > Ganesh Raja <ganesh.raja@xxxxxxxxx> wrote: > Nisar, > > Did not Understand too much from your Post .. But How does all the > things that you have talked about affect the way parallelism works .. > Parallelism is CPU Bound and not disk bound ... U dictate if you want > to Use Parallelism and then The ur Query/DML is Split across multiple > processess that works out how to fetch data from the database and > processes it. > > I am disagreeing here with irfhan also when he says that Multiple > extents will give u chances of better parallelism. U can write a small > testcase with a One Extent Table and See what happens .. U will still > get Parallel Reads. > > The Only thing that has some effect on performance with Multiple > Extent is the MBRC value since an Multi Block read cannot be done over > multiple extents. [ U will never encounter this has a Problem] > > Apart from that i will never say that if you have 1000 Extents ur > Query is Going to Perform badly and if you have 1 Extent it will > perform any better. I need a test case from u to prove that the > otherway. > > Extent has always been a Logical Grouping of Blocks and not a > Physicall Grouping. > > My $0.02 and not my logic ... :) > > Rgds > Ganesh > > --------------------------------- > Do you Yahoo!? > Check out the new Yahoo! Front Page. www.yahoo.com > >