Oracle FAQ Your Portal to the Oracle Knowledge Grid

Home -> Community -> Mailing Lists -> Oracle-L -> RE: Oracle replication with EMC2

RE: Oracle replication with EMC2

From: Hameed, Amir <>
Date: Fri, 21 Jul 2006 08:44:18 -0400
Message-ID: <>

We have been using the BCV split method with Oracle hot backups for the last few years and it has been working fine. We use TimeFinder's symioctl utility to put the database in hot backup mode, which essentially issues Oracle's begin ..end commands under the hood. We are not using consistent splits where the IO is frozen via powerpath. I believe I had found an issue during testing with this approach when doing incomplete recovery.

[] On Behalf Of Zanen, dhr. J.A. (Jack) van

	Sent: Friday, July 21, 2006 2:33 AM
	Subject: RE: Oracle replication with EMC2



	We had a lengthy discussion about this at a previous site where
the person in charge of wheeling this in refused to see our point or contact the vendor.

        Our backup strategy revolved around BCV splits that we, the DBA team, did not approve of.

        Above mentioned person knew better and refused to listen to all our concerns and advise. We had to get the vendor to send him some documents to point out to him that if you planned to use BCV for backup you'd have to put the tablespaces in backup mode.          

        BCV's are restartable images, I believe is what they call these BCV splits. (been a while)                    


[] Namens Matthew Zito

	Verzonden: vrijdag 21 juli 2006 1:58
	Onderwerp: RE: Oracle replication with EMC2



	All a consistent split does, on the symmetrix, is freeze IOs to
the volumes while the split is in progress. This is because a BCV relationship is like this:         
	Lun 1 - M1-M2--M3
	Lun 2 - M1-M2--M3
	Lun 3 - M1-M2--M3
	Where each M is a physical disk.  When you split them, its like
a zipper pulling down:         
	Lun 1 - M1-M2-  -M3 (split completed)
	Lun 2 - M1-M2- -M3 (split in progress)
	Lun 3 - M1-M2--M3 (not split yet)
	So, if you're striping with, say, a volume manager across those
three luns, you could have a situation where you do a long write that covers all three luns, and since the split was in progress, one of the BCV disks got the write and the others didn't, resulting in inconsistent data.         

        The consistent split freezes the IOs, so that its impossible for a write to go to a disk that hasn't split yet. However, to the database, this still looks like a shutdown abort, or a server crash. EMC still recommends doing the begin backup before you do this.         

	Matthew Zito
	Chief Scientist
	GridApp Systems
	P: 646-452-4090
	-----Original Message-----
	From: on behalf of Riyaj
	Sent: Thu 7/20/2006 2:55 PM
	Subject: Re: Oracle replication with EMC2
	 >>When they do the bcv mirror split, they don't inform Oracle
	tablespaces in hot backup mode), so it seems these files will
not be
	 >> consistent for what  Oracle expects.
	It is possible to have consistent backup without altering the
	tablespaces in to backup mode, using a special type of split
known as
	consistent BCV splits. This method is supported by EMC for
	database backups and mentioned in various documents. You need
	or ECA for the consistent split to work. I/O to that device is
	while the split is underway.
	Even then, I would be inclined to alter the tablespaces to
backup mode
	and then split.
	Yechiel Adar wrote:

>> <.. snipped...>
>> When they do the bcv mirror split, they don't inform Oracle (put >> tablespaces in hot backup mode), so it seems these files will not be >> consistent for what Oracle expects.
>> <.. snipped ..>
>> > -- > > > > ________________________________ Aan dit bericht en eventuele bijlagen kunnen geen rechten worden

        Het Provinciaal Bestuur van Noord-Holland.         

Received on Fri Jul 21 2006 - 07:44:18 CDT

Original text of this message