Re: i/o parameters on oracle and sun

jonathon and connor have both already hit the (proverbial) nail on the head. and, at a risk of delivering a glancing blow and bending it over sideways... here's my swing at it

assuming an oracle block size of 4K

the odds that a single multiblock read of 16 blocks (64K) will require i/o operation on more than one device are:

  1 in 2 given a stripe size of 128K
 15 in 16 given a stripe size of 64K
  1 in 1 given a stripe size of 32K

the odds that a single multiblock read of 8 blocks (32K) will require i/o operations on more than one device are:

  1 in 4 given a stripe size of 128K
  1 in 2 given a stripe size of 64K
  7 in 8 given a stripe size of 32K

the odds that a single multiblock read will require more than one i/o operation on a single device are:

  1 in 1 if request size > stripe size * no.of disks   0 if request size <= stripe size * no.of disks

the odds that multiple i/o operations against a single device will be combined together into a single larger i/o operation (e.g. by async or list io) are:

  dependent on factors in addition to stripe size.

the odds that a block will already be in the buffer cache and not require any i/o operation are:

  dependent on factors other than stripe size.

does anyone care to place a wager ?

"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote in message news:979814882.27443.0.nnrp-02.9e984b29_at_news.demon.co.uk...
>
> More particularly, you do not want a single multiblock read
> to be larger than the amount of data that can be found on a
> single disk (unless the system is a single user system) otherwise
> a single Oracle read request ends up queuing and blocking on more
> than one device.
>
> --
> Jonathan Lewis
> Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk
>
> Practical Oracle 8i: Building Efficient Databases
>
> Publishers: Addison-Wesley
> See a first review at:
> http://www.ixora.com.au/resources/index.htm#practical_8i
> More reviews at: http://www.jlcomp.demon.co.uk/book_rev.html
>
> Connor McDonald wrote in message <3A643625.979_at_yahoo.com>...
> >thus smaller stripe sizes could perform better - but as you said, you
> >would not want a single multblock read to be larger than no. disks *
> >stripe size since you would get nasty 'wraparound'.
> >
>
Received on Thu Jan 18 2001 - 22:06:42 CST