Rman copy process

[Peter Day_3]

Guys,
I recently had to recreate a filesystem that is used by the DBA's for doing a 'rman' hotbackup.

The problem I am now getting is that the Oracle blocksizes have been set to 16Mb, and when I recreated the new fileystem I didn't specify any blocksizing. This rman process is now taking a lot longer, why, and how can I check the block sizing when it is set.

Thanks,
Pete

[Robert-Jan Goossens · Accepted solution]

Hi Pete,

Check the block size with,

# fstyp -v /dev/vg01/lvol1

# df -g /app/oracle/admin

Regards,
Robert-Jan

[G. Vrijhoeven]

Hi Pete,

To see the command used to create a filesystem you can use the mkfs with the -m option and the -F option. ( check man mkfs_vxfs) The default is 1024.


mkfs -F vxfs -m /dev/dsk/c?t?d? ( /dev/vg??/lvol?)

HTH,

Gideon

[Nicolas Dumeige]

Hello,

"Oracle blocksizes have been set to 16Mb"
16 Mb or 16 k ?

Cheers

Nicolas

[A. Clay Stephenson]

If this is a vxfs filesystem (and it should be because vxfs filesystems have outperformed hfs for several years now) believe it or not, block size (filesystem that is) has almost zero impact on performance. Why? Because vxfs filesystems, unlike hfs, are extent-based and the filesystem block size really has no impact on performance. "Database block size should equal filesystem block size" is one of those truism's that has been beaten into DBA's heads but the real answer is that it depends upon the type of filesystem.

[Stephen Andreassend]

We've found that the vxfs block size does make a difference on performance on really slow disks. We were ftping files and found that the IO write rate went from 2MB/sec to 8MB/sec by increasing it from 1K to 8K. However we've also found that for write intensive Oracle processing, doesnt really have that much of an effect.

[Stephen Andreassend]

You may be able to increase the RMAN speed by increasing the db_block_size (requires DB rebuild), or DB_FILE_DIRECT_IO_COUNT. In Oracle 9i it's _DB_FILE_DIRECT_IO_COUNT.

Also you can set the readrate parameter in RMAN:

"specifies the maximum number of buffers (each of size DB_BLOCKSIZE * DB_FILE_DIRECT_IO_COUNT) per second read for backup or copy operations from each of the input datafiles. By default, this parameter is not set. Use this parameter to "throttle back" RMAN, that is, set an upper limit for block reads so that RMAN does not consume excessive disk bandwidth and thereby degrade online performance."

May be set that to a high value?

setlimit channel c1 readrate 2000;