Fwd: Lamport vs Broadcast on Commit - RAC only
- From: K Gopalakrishnan <kaygopal@xxxxxxxxx>
- To: "oracle-l@xxxxxxxxxxxxx" <oracle-l@xxxxxxxxxxxxx>
- Date: Wed, 27 Jul 2005 15:36:37 +0530
---------- Forwarded message ----------
From: Anand Rao <panandrao@xxxxxxxxx>
Date: Jul 27, 2005 3:29 PM
Subject: Re: Lamport vs Broadcast on Commit - RAC only
To: racdba@xxxxxxxxxxxxx
Hi Brian,
Yes, as KG mentioned, Tru64 is an exception in the SCN scheme of
things. Also, the version and platform is crucial.
It really depends on the rate of changes in your application. How many
changes are you generating per second? changes mean
(UPDATE/DELETE/INSERT or any DDL) per second PER NODE.
I frankly don't think interconnect bandwidth is a factor for a 24-byte
commit scn structure to pass through.
Sometimes, Oracle can piggy back the Commit SCN along with other
messages which could be delay of a few milliseconds. But, 1000 ms
makes 1 second ! so, you really need to have an approximate (nearly
correct) idea of your COMMIT rate.
corrections welcome,
regards
anand
On 27/07/05, K Gopalakrishnan <kaygopal@xxxxxxxxx> wrote:
> Brian:
>
> You have not mentioned the version and platform. So I assume it is
> 9iR2 and some flavor of UNIX other than Tru64. Please correct me
> otherwise, as the results vary on 10gR2 and Tru64.
>
> Broadcast on Commit is certainly expensive. But I have not seen very
> noticeable performance overhead on BOC in our benchmarks. Others say
> the overhead is around 5-10%. Recently I had seen one case where the
> top wait was 'wait for scn from all nodes' when BOC is used. You
> don't need to worry much unless you are seeing huge waits for 'wait
> for scn' wait. Packaged Applications (Oracle Ebusiness suite or SAP)
> require MCPD as 0 to force BOC SCN propagation.
>
> Starting from 10gR2 BOC is the default method for SCN propagation as
> they have change LGWR to asynchronous.
>
> CC: racdba-s: Do you guys have any comments?
>
>
> Regards,
> Gopal
>
>
>
> On 7/26/05, Brian Wisniewski <brian_wisniewski@xxxxxxxxx> wrote:
> > Has anyone done testing using the lamport vs broadcast on commit scheme and
> > can comment on the results?
> >
> > We have an application where we're needing to use the broadcast on commit
> > due to the way the application was designed. I'm curious as to what type of
> > performance hit will be associated with it. Oracle's defines the hit as
> > "marginally more resource intensive" but somehow I'm not all warm and fuzzy
> > with that definition.
> >
> > I'm want to run some tests to try and determine the impact but just don't
> > have the time right this moment. I'd appreciate feedback from anyone who's
> > tested this and would be willing to share their results.
> >
> > Thanks - Brian
> >
> >
> >
> >
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam? Yahoo! Mail has the best spam protection around
> > http://mail.yahoo.com
>
>
> --
> Best Regards,
> K Gopalakrishnan
> Co-Author: Oracle Wait Interface, Oracle Press 2004
> http://www.amazon.com/exec/obidos/tg/detail/-/007222729X/
>
>
--
Best Regards,
K Gopalakrishnan
Co-Author: Oracle Wait Interface, Oracle Press 2004
http://www.amazon.com/exec/obidos/tg/detail/-/007222729X/
--
//www.freelists.org/webpage/oracle-l
Other related posts:
- » Fwd: Lamport vs Broadcast on Commit - RAC only
