RE: Optimized redo log access
- From: "Mark W. Farnham" <mwf@xxxxxxxx>
- To: <agonenil@xxxxxxxxx>, "'K Gopalakrishnan'" <kaygopal@xxxxxxxxx>
- Date: Mon, 22 Sep 2014 09:31:35 -0400
So if you use external redundancy you are left to whatever feature set is
contained in the volume manager controlling your storage.
When using external redundancy the storage is presented to Oracle as a single
logical entity, so there are not alternative plex handles for which Oracle
could set a preference.
Within the black box you present to Oracle as external redundancy there may or
may not be controls depending on the functionality of that volume manager. If
you have built volumes with mixed loading and capabilities within an external
volume manager, that is something I don’t think Oracle should even try to
manipulate. Rather, that is for either the DBA or the DBA, network, and storage
teams together to work out to best advantage.
Likewise, when you are managing everything with ASM, this is why it is usually
most productive to give Oracle “luns” that are whole single disks or whole
single trays to managed without attempting manipulations at the native volume
manager level. Essentially you either manage nothing but the naming and
allocations to disk groups or you manage everything in ASM. Very likely if you
try to manage these things from both sides simultaneously (often by two or more
distinct persons) you will end up working at cross purposes.
Of course convincing storage teams and network teams why things allocated to
Oracle shouldn’t be used even intermittently for other or competing purposes
has always been the tricky part.
mwf
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On
Behalf Of amihay gonen
Sent: Monday, September 22, 2014 3:08 AM
To: K Gopalakrishnan
Cc: Mark W F; Jeremiah Wilton; xiangdongzou@xxxxxxxxx; ORACLE-L
Subject: Re: Optimized redo log access
this feature is applicable only to normal or high redundancy and not external
redundancy :(.
On Mon, Aug 18, 2014 at 4:11 AM, K Gopalakrishnan <kaygopal@xxxxxxxxx> wrote:
All--
Starting from 12.1 , ASM "Even Read" feature automatically sends the read
request to least loaded disk groups. So if the OP's objective is to distribute
evenly , then the functionality is already in the kernel.
-Gopal
On Sun, Aug 17, 2014 at 7:43 PM, Mark W. Farnham <mwf@xxxxxxxx> wrote:
yeah, ping ponging online redo groups was standard operating procedure back in
the day.
I think the person asking has members on disks with dramatically differing
loads, so he wants arch to use the member on the less busy drive.
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On
Behalf Of Jeremiah Wilton
Sent: Sunday, August 17, 2014 4:12 PM
To: Mark W. Farnham; xiangdongzou@xxxxxxxxx; 'agonenil'; 'ORACLE-L'
Subject: RE: Optimized redo log access
Amihay pointed out that I probably meant group instead of member.
JW
Sent from my Verizon Wireless 4G LTE smartphone
-------- Original message --------
From: Jeremiah Wilton
Date:08/17/2014 12:19 PM (GMT-08:00)
To: "Mark W. Farnham" , xiangdongzou@xxxxxxxxx, 'agonenil' , 'ORACLE-L'
Subject: RE: Optimized redo log access
Back when it mattered which disk device we accessed for a specific operation,
this was pretty easy to achieve.
You just alternate disk(s) by member. For instance, if you have two dedicated
disks for online logs, member 1 would be on disk A and member 2 on disk B. That
way when we switch from member 1 to member 2, the archiver reads from disk 1
while we write to the online log on disk 2. You can see easily how we could
even do this and keep multiplexing with 4 disks.
I doubt that you get much benefit trying to do this kind of thing on modern
storage.
Jeremiah
Sent from my Verizon Wireless 4G LTE smartphone
-------- Original message --------
From: "Mark W. Farnham"
Date:08/17/2014 6:13 AM (GMT-08:00)
To: xiangdongzou@xxxxxxxxx, 'agonenil' , 'ORACLE-L'
Subject: RE: Optimized redo log access
I also do not know a way to make the archiver read a preferred member.
However, you may be able to get the read from a preferred physical disk
depending on your volume manager.
In ASM this would be by setting (or your preference being the default) the asm
preferred read failure group as shown in this example from the manual so that
your preferred read is the first one:
alter system set asm_preferred_read_failure_groups =
'DG2.DG2_0000','DG3.DG3_0000'
In some other volume managers there are possibilities for influencing or
controlling the preferred read of a volume constructed from plexes on different
physical media.
The starting point recommendation for multiple members is they should be on
media of similar capabilities and load, so the archiver (unless they slipped it
in when I wasn’t paying attention) does not directly have a toggle for
preference by member.)
Good luck. (Of course feeding arch is the reason some folks put redo on SSD,
while benchmarkers run around “debunking” the idea of redo on SSD because they
can show it does not speed up writes to compared to isolated physical HD
volumes when they compare to flash SSD [not battery backed DRAM SSD] with arch
turned off.) Removing arch reads from your diskfarm by relocating redo to SSD
will in fact “deheat” your disk farm and remove the competition for seeks for
writes and bandwidth for arch reads. Since redo is usually small relative to
the database disk farm, this may be cost effective, depending on your situation.
mwf
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On
Behalf Of xiangdongzou
Sent: Sunday, August 17, 2014 5:27 AM
To: agonenil; ORACLE-L
Subject: 回复: Optimized redo log access
archiver process can't read only special member.Oracle decicde which
member shoulde read.
2014-08-17
_____
I AM AN ORACLE FANS!
Skype:Frank.oracle
Email:xiangdongzou@xxxxxxxxx <mailto:Email%3Axiangdongzou@xxxxxxxxx>
_____
发件人：amihay gonen <agonenil@xxxxxxxxx>
发送时间：2014-08-17 16:51
主题：Optimized redo log access
收件人："ORACLE-L"<oracle-l@xxxxxxxxxxxxx>
抄送：
Hi , I've system that have multiplex redo log (2 members per group) and
multiplex destinations .
Due to performance consideration I would like to configuration the archiver
process to read only from one member (to reduce load on the specific device).
I'm not aware of such option to tell oracle which member to read .
Does anyone know if it is possible ?
Amihay .
Other related posts:
