Re: RAC setting for max consistency?

From: Robert Klemme <shortcutter_at_googlemail.com>
Date: Thu, 13 Sep 2007 19:34:55 +0200
Message-ID: <5ktam2F5encqU1@mid.individual.net>

On 12.09.2007 18:56, joeNOSPAM_at_BEA.com wrote:
> On Sep 11, 2:23 pm, Robert Klemme <shortcut..._at_googlemail.com> wrote:

>> On 11.09.2007 20:08, joeNOS..._at_BEA.com wrote:
>>
>>
>>
>>> On Sep 11, 8:56 am, "joeNOS..._at_BEA.com" <joe.weinst..._at_gmail.com>
>>> wrote:
>>>> Hi all,
>>>> IIRC, there is a RAC configuration setting that determines the level
>>>> of consistency in the view
>>>> of the data between multiple RAC nodes. The tradeoff is between
>>>> maximum consistency and
>>>> coordination versus an acceptance of some latency of view allowing
>>>> faster performance. Can
>>>> you tell me this configuration parameter and the setting for maximum
>>>> consistency?
>>>> thanks much,
>>>> Joe Weinstein
>>> I found it: MAX_COMMIT_PROPAGATION_DELAY
>>> Under some circumstances with rapid writes and reads, the data
>>> seen at node B may not include committed changes just made by
>>> node A. This setting is for 10.1 and earlier.
>>> Joe Weinstein at BEA Systems
>> For distributed TX there is also tight and loose coupling which also
>> influences visibility of changes among TX branches:
>>
>> http://download.oracle.com/docs/cd/B19306_01/appdev.102/b14251/adfns_...
>>
>> Kind regards
>>
>>         robert

>
> Thanks! Interesting:
>
> Tight and Loose Coupling
>
> Application threads are tightly coupled if the RM considers them as a
> single entity for all isolation semantic purposes.
>
> Do you know where it is documented how the RM decides if threads are a
> single entity for these purposes?
> Would it simply be the TX ID associated with the threads? (different
> branches maybe)

I haven't worked with RAC so far. But I know that the application deployer of a JEE application can decide whether an Oracle XA connection uses tight or loose coupling - so I think this is the user's decision. As far as I understand, Oracle does extra work if it detects different branches of the same TX that are tightly coupled. In those cases they can see their respective chances and probably also do not lock each other out of resources. It seems loosely coupled branches just work like two ordinary TX and the only difference is that you can commit them via 2PC.

There's probably more information in the DT standard of XOpen (now Open Group). I remember having seen those terms in that doc (a PDF). I don't have the URL handy but I can look it up if you are interested.

Kind regards

        robert Received on Thu Sep 13 2007 - 12:34:55 CDT