How about..
A reader, June 05, 2013 - 2:01 am UTC
if your databases are hosted on virtual machines? Do you still think it's necessary to have the redos and control files multiplexed?
June 05, 2013 - 1:19 pm UTC
why would a virtual machine change anything? what feature/function of a virtual machine do you think would be different than a non-VM environment regarding this?
In short "yes"
Follow-up
A reader, June 05, 2013 - 2:44 am UTC
"I would have mirrored redo - and if at all possible, have that redo being transmitted as soon as possible to a standby site via data guard if the data is important. "
Let's say we can't afford to pay an extra 100% on Oracle licenses to implement a standby site, would you still create multiple redo log members if the logs are protected on a SAN?
I know that it depends.
Here are a few scenarios. Please state whether you would still add redo log members which are protected by SAN:
1) we have a high amount of database change and we can afford to lose all the data. Yes, we can afford to lose all the data. Don't ask why this is the case.
2) We have a high amount of database changes and we can lose data, but the customer won't be happy. However, customer will fire me if the extra redo log members slow performance down enough that we don't meet SLAs.
3) We have a low amount of database changes.
June 05, 2013 - 1:23 pm UTC
would you still create multiple redo log members if the logs
are protected on a SAN?
As written: "I would have mirrored redo - and if at all possible,, I would definitely have mirrored redo.
and if possible - a failover if the data is important.
that wasn't an "it depends", that was a "I would have mirrored redo".
1) can afford to <> really want to. So, therefore, I would have mirrored redo/control files.
2) see #1 (mirrored redo isn't going to slow performance, the writes are in parallel)
3) see #1
Clarification Request
John Cantu, June 07, 2013 - 1:54 am UTC
Regarding your statement: "2) see #1 (mirrored redo isn't going to slow performance, the writes are in parallel) "
I am quite confused regarding the fact that you stated that since it is written in parallel, it can't cause a I/O performance hit. With two redo log members per redo log groups, the system now has to write to two different files on the filesystem, so there should be 2X the throughput that the storage unit must process. Right?
June 07, 2013 - 1:33 pm UTC
if you put your MIRRORED redo on the same devices, you are doing it wrong.
you want to use physically separate things.
confused
A reader, July 04, 2013 - 6:22 am UTC
Tom,
You mentioned here (
http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:8400628663041#5621977000346829549 ) that you don't do OFA and just accept the default configuration of the installer out of the box.Considering the kind of setup you also suggested for the questioner there (just one big mount point e.g. /u01),
1. Wouldn't that mean that redo and control files are actually placed only on subtrees of the mount point /u01 and not actually on a separate disk/device?
2. If so, wouldn't this mean that these files are not actually protected at all?
3. If the answer to #2 is yes, should the DBA then relocate the said files to a different mount point?
Thank you for enlightening us about the Oracle database. Thanks.
July 16, 2013 - 1:11 pm UTC
1) assume striped file system. assumed "normal" system - not high end OLTP. then yes, one big mount point is fine. It is what we default to in ASM too. Just strip and mirror everything.
2) the disk array should protect them from physical loss, yes, but we still need to protect them from logical errors and human errors.
3) depends on the nature of the underlying disk array entirely. Assuming the 21st century and assuming they would be striped and mirrored, it may typically not be necessary. the storage is doing redundancy for disk loss protected, we do redundancy for logical corruption and human errors.