LVM steps for copying oracle raw volumes, to another server.

[rveri]

Hi Experts,

1) What is the best way to copy the oracle raw logical volumes , for migrating to another host.

Suppose I have 100 raw lvol:
- /dev/vg_OR11/rr_lvol1 to rr_lvol100

How to copy it to another server with another storage.


2) If doing dd, is it required oracle to be down during dd.


3) what dd command to be use, block size etc.
Thanks,

[TTr]

1) You can use dump-and-load or dd. Dump-and-load is mostly DBA work, CPU and disk i/o resources, the disks on the new server do not have to be identical to the disks on the old server. Using dd is mostly unix admin work, i/o and LAN resources, the disks on the new server must be exactly the same (name and size) as those on the old server.

2) Yes with dd oracle MUST be down, absolutely no disk writing activity.

With either method, you need to have oracle down to capture all the oracle disk changes up to the last moment.

With the dump and load method, you may be able to leave oracle UP but you MUST utilize the latest archive logs (after the dump moment) from the old server to update the database on new server.

3) The block size is optional but if you use it it must be the same on both servers. You may want to do some tests to find out which block size gives you better i/o results.

Something like this would work (you need to allow remsh from the old to the new server):

dd if=/dev/vgXX/rlvYY | remsh newserver "dd of=/dev/vgXX/rlvYY"

[Bill Hassell · Accepted solution]

> 3) The block size is optional but if you use it it must be the same on both servers. You may want to do some tests to find out which block size gives you better i/o results.

Not really optional unless you don't care how many days it will take to copy several GB. The default is 512 bytes. If you don't set bs to a reasonable value, the copy time will be as much as 100x longer! I would recommend the option:

bs=1024k

which means each read and write will be a megabyte. If you have some spare time, try some other bs= values. The actual speed is VERY dependent on whether you are using slow SCSI disks, fibre disks (JBOD) or a disk array (HP, EMC, Axiom, etc). You can try different bs= sizes. Be sure to use count= so the amount of data transferred is the same:

time dd if=... of=... bs=512k count=200
time dd if=... of=... bs=1024k count=100
time dd if=... of=... bs=2048k count=50

[Jean-Luc Oudart]

Hi

you did not mention the source and target machine and OS / OS version, nor Oracle version.

Another option is to use RMAN (Oracle Recovery manager)
see doc link for 9i
http://download.oracle.com/docs/cd/B10501_01/server.920/a96566/rcmdupdb.htm#441628

Regards
Jean-Luc