Disk Arrays
cancel
Showing results for 
Search instead for 
Did you mean: 

ORACLE, JFS, and FC10 array

SOLVED
Go to solution
George Nikoloudis_1
Frequent Advisor

ORACLE, JFS, and FC10 array

Dear all I was reading the Fibre Channel protocol specifications and there I found that the payload of the Fibre Channel Protocol is 2148 Byte. In my system I have JFS with block size of 8192 and fragment size of 2048. What I found out is that the JFS reads/writes 64 KB chunks no matter if the block size 1,2,4 or 8 KB The Oracel DB has db_block size of 8192. What I understand is that each time the Oracle asks for a new data then the UX performs 8 reads for geting 1 block (8192) ie 64 reads for the 64 KB for JFS. Oracle needs only 8KB
The above are True? Are there any problem in the perfromance?
My configuration is a V CLass 8Process/8GB Memory, FC10 over HUB.
7 REPLIES
Roger Baptiste
Honored Contributor

Re: ORACLE, JFS, and FC10 array



hi, what command/output are you using to determine this?? Please note that the "fragment size" displayed by df -g is
the actual block size. This can also be seen with the fsadm command.

HTH
raj
Take it easy.
Duncan Edmonstone
Honored Contributor

Re: ORACLE, JFS, and FC10 array

Hmmm - you need to be careful about what your block size really is - did you get your info from fstyp -v ? A better way of determining your block size for VxFS file systems is to do:
mkfs -m /dev/vgXX/lvolYY

And look at the bsize value.

Don't worry aboyt what VxfS is doing behind the scenes, or about the FC packet payload size. Just make sure FS Block size=Oracle block size

If you have Online JFS and the server is an Oracle DB server only, you can also tune file system buffer caching to a nice low value:

Set dbc_max_pct and dbc_min_pct to low value (e.g. 5 for both)
Disable file system buffer caching for the VxFS file systems by mounting with options 'mincache=direct,convosync=direct'
Increase the size of your Oracle buffer cache to take advantage of the freed up memory.

HTH

Duncan

HTH

Duncan
Alexander M. Ermes
Honored Contributor

Re: ORACLE, JFS, and FC10 array

Hi there.
Just forget about the chunks and stuff.
Set your OS blocksize to 8k and the Oracle block size to the same value. That will do perfect. We run these values on a V2500 with an EMC Symmetrix with FC interfaces without performance problems.
Your db_block_buffers value depends on the size of the database and the user activity.
You can check it, if you go for the library cache hit ratio ( Oracle Enterprise Manager / Performance Manager ).
Rgds
Alexander M. Ermes
.. and all these memories are going to vanish like tears in the rain! final words from Rutger Hauer in "Blade Runner"
George Nikoloudis_1
Frequent Advisor

Re: ORACLE, JFS, and FC10 array

Thanks for your reply.
I run the command
fstyp -v /dev/vg01/mnt1a and the output was:
vxfs
version:3
f_bsize:8192
f_frsize:2048
What is f_bsize and f_frsize ?
Which one corresponds to the block size?
Can I change the block size?

Thanks
George Nikoloudis_1
Frequent Advisor

Re: ORACLE, JFS, and FC10 array

How can I mount my filesystems with different parameters, Please note that I am using MC/Service Guard.

Thanks a lot
Duncan Edmonstone
Honored Contributor
Solution

Re: ORACLE, JFS, and FC10 array

Your FS block size is 2048.

No you can't change the block size unless you re-create the file system.

THe mount options for file systems in ServiceGuard go in your package control script, these should be in a sub-directory under /etc/cmcluster (if your not sure which use cmgetconf to determine which is your package control script). FInd the lines that look something like this:

LV[0]="/dev/vgXX/lvolYY"; FS[0]="/u01"; FS_MOUNT_OPT[0]=""
LV[1]="/dev/vgXX/lvolYY"; FS[1]="/u02"; FS_MOUNT_OPT[1]=""

and change the FS_MOUNT_OPT parameters to match the options you want... e.g.
FS_MOUNT_OPT[0]="-o largefiles,mincache=direct,convosync=direct"

HTH

Duncan

HTH

Duncan
Duncan Edmonstone
Honored Contributor

Re: ORACLE, JFS, and FC10 array

Should also point out that you need to distribute your package control script to all nodes in the cluster if you change it.

HTH

Duncan