Simpler Navigation for Servers and Operating Systems
Completed: a much simpler Servers and Operating Systems section of the Community. We combined many of the older boards, so you won't have to click through so many levels to get at the information you need. Check the consolidated boards here as many sub-forums are now single boards.
Showing results for 
Search instead for 
Did you mean: 

LVM IO sizes and Oracle writes

Go to solution
Stephen Andreassend
Regular Advisor

LVM IO sizes and Oracle writes

For file systems, I believe the max LVM IO size is 256K using file systems or 1MByte using asynchronous IO.

If I were to set Oracle redo log write sizes to match the max LVM IO size, rather than 1 big IO clump when it flushes the redo log buffer, would I necessarily get a faster write?Why? Surely doing just 1 IO call is faster than doing several evenly sized IO calls that match the IO size limit?
Stephen Andreassend
Regular Advisor

Re: LVM IO sizes and Oracle writes

In otherwords, will doing multiple IOs equal to the max LVM IO size be faster than doing 1 big IO from an application point of view?

ie Is there a real performance issue here or is this just academic?
Bill Thorsteinson
Honored Contributor

Re: LVM IO sizes and Oracle writes

IO that large will likely be chunked by the driver anyway.
HPUX sorts the IO requests by
location. Modern disks
don't have real HCS mapping
visible on the interface.
Choosing the max LVM IO size
will likely block all other
IO on the disk while it completes. This may be
slightly faster than smaller

If your IO request is too large you may block other
IO on the same disk. This
could slow overall performance.

Unless you devote the disk to
redo logs, I would expect the
gains to be acedemic. Do you
expect to block waiting for
redo writes? Do you expect
to have the disk fully loaded
with IO requests.

Acedemically, you are deep into queuing theory.

Stephen Andreassend
Regular Advisor

Re: LVM IO sizes and Oracle writes


I guess I need a white paper on performance tuning queuing in the LVM.