RE: "wait for scn from all nodes" wait in RAC
- From: "Hameed, Amir" <Amir.Hameed@xxxxxxxxx>
- To: "K Gopalakrishnan" <kaygopal@xxxxxxxxx>
- Date: Sat, 16 Jul 2005 12:46:40 -0400
KG,
Thank you for the explanation. The MCPD is set to 0, a recommendation
from Oracle for the 11i databases configured for RAC. Am I correct to
conclude that because this is a synchronous process, the speed of the
interconnect plays an important role in this activity and the more nodes
you have in RAC the more will the impact? We are using three gigabit
private interconnects between the nodes. The Veritas LMX module also
multiplexes the traffic over interconnects, so I am assuming that in
theory, our interconnect speed is fine and should not be a bottleneck.
But what else could cause this process to slow down and show up as among
the top five?
-----Original Message-----
From: K Gopalakrishnan [mailto:kaygopal@xxxxxxxxx]
Sent: Saturday, July 16, 2005 7:15 AM
To: Hameed, Amir
Cc: oracle-l@xxxxxxxxxxxxx
Subject: Re: "wait for scn from all nodes" wait in RAC
Amir:
What is the value of the MPCD settings? The scn propagation algorithms
are based on the max_commit_propagation_delay parameter and the values
are platform dependent. And from your wait I assume you have set the
MCPD is <>700 cs and you may be using 'broadcast on commit' method for
SCN propagation across the nodes.
In Broadcast SCN method, on every commit we propagate the SCN to all
the nodes, before writing the redo, we wait for the response from all
the nodes and then complete the redo. And all the processes will have
to wait till we write the redo to the disk. THis is synchronous in 9i.
There are some enhancements in 10g on this area and the log writing
process is asynchronous to SCN propagation. This is controlled by
_lgwr_async_broadcasts parameter which defaults to true. In this case,
if we receive any SCN > commit SCN we don't need to broadcast (and
wait ) our SCN.
On 7/16/05, Hameed, Amir <Amir.Hameed@xxxxxxxxx> wrote:
>
>
> Does anyone know what this wait (wait for scn from all nodes) is about
in
> RAC. We are currently running a test on a two-node RAC/11i instance.
> Statspack is showing this wait among the top five (at #5). We are
using the
> "broadcast on commit" scheme as it is recommended by Oracle for 11i
> instances. Per Metalink, this event is related to the broadcast scheme
that
> we are using. But what does it really mean and at what point it
becomes a
> matter of concern? Any feedback will be appreciated.
>
> Thank you
> Amir
--
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:
