Changing max_io_size parameter oracle

[ricardo_24]

Hi,

I need change "max_io_size" oracle parameter to 8192000, but, i know this value is in kernel table. In kernel, dont have a similar name, the more apear is "scsi_maxphys". Any know as do to change this value?

Tanks,

Ricardo

[G. Vrijhoeven]

Hi Ricardo,

Check:
http://docs.hp.com/hpux/onlinedocs/939/KCParms/KCparams.OverviewAll.html

Gideon

[Jim Butler]

Oracle tunables are
dbc_max_pct (should evaluate to 128 Mb)

maxdsiz
maxssiz

and the semaphore entries are
semmnu
semmns
semume (and about 3 others)

Shared memory is shmmax -

[Jim Butler]

Oracle tunables are
dbc_max_pct (should evaluate to 128 Mb)

maxdsiz
maxssiz

and the semaphore entries are
semmnu
semmns
semume (and about 3 others)

Shared memory is shmmni and other shm...
entries

I do not believe you want to change the scsi_maxphys unless you have more than that amount of physical scsi chains on the system.

[Stuart Abramson_2]

There is no UNIX kernel parameter for "max_io_size".

On "newfs" command, there is a "-b" parameter which changes block size for i/o to a mounted file system. I don't think it goes that big. Do "man newfs_vxfs".

[ricardo_24]

Friends,

I search this parameter in metalink oracle e find a note coments about this case, is write is no a lvm parameter on HP-UX, the max size is 1 mb by default.

Thanks for help.

Ricardo

[Stuart Urquhart]

I think this may be a JFS paramter. Create a file /etc/tunefstab with a line like:

/dev/$VG/lvol$n max_io_size=8192000

for each filesystem/lvol you want to change
and run

/sbin/vxtunefs -s -f /etc/tunefstab

I think you might need an OnLine JFS license for this to work.

[Todd McDaniel_1]

Ricardo,

AS you found out there is no max_io_size parm in the kernel...

If this is in reference for Oracle, then you need not worry about increasing any such parm

Oracle works in blocks...4k, 8k, 16k, 32k, etc...

HPUX works in extents...4mb minimum and in multiples of 4mb, usu up to 256MB for striping...

Extents can be/are many times larger than any block size that Oracle can work in even up to 32k blocks.

When dealing with Oracle and the HP Kernel, most of the parms that affect it are as listed above...

[Sridhar Bhaskarla]

Hi,

There is no kernel parameter called max_io_size.

If you are using raw logical volumes, the maximum IO will be 1 megabyte. On a filesystem it is 256k.

But you can tune it if you have JFS3.3. YOu can change read_pref_io and write_pref_io to change this though they are more helpful if you have striping.

Look at 'man vxtunefs' command.

-Sri