| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Mailing Lists -> Oracle-L -> RE: RAC 10g interconnect resilient configuration
Luca,
I'm glad to help.
As for the VIP, I meant to recommend bonding the public interfaces also due to this bug.
As for the dual switches, all servers connect to both switches. One NIC in each bonded pair connects to each switch. In this way you have redundancy in the NICs and the switches. Make sure and turn spanning-tree off on the switches for the private network.
If your private network goes down, you lose all nodes except the primary node.
Bryan
-----Original Message-----
From: Luca Canali [mailto:Luca.Canali_at_cern.ch]
Sent: Monday, February 20, 2006 3:43 PM
To: bthomas_at_perftuning.com; oracle-l_at_freelists.org
Subject: RE: RAC 10g interconnect resilient configuration
Hi Bryan,
Thank you for the link. Although I have one main remark with the architecture they propose: the system scalability looks to me (see page 10) limited by inter-switch communication on the private network (they can have clusters with half of the nodes on one switch while the other half is connected to the other switch).
As for your comment, I am not sure I can see how the interconnect bonding configuration can influence the VIP (that is public network) fail over behavior, unless you're hitting some strange bug.
Thanks,
L.
-----Original Message-----
From: Bryan Thomas [mailto:bthomas_at_perftuning.com]
Sent: Monday, February 20, 2006 10:06 PM
To: Luca Canali; oracle-l_at_freelists.org
Subject: RE: RAC 10g interconnect resilient configuration
Luca,
I would highly recommend anyone setting up RAC to using bonding for the interconnect.
We have seen problems with the VIP not failing over properly in certain circumstances.
Dell has a good document that describes the NIC bonding located here:
http://support.dell.com/support/edocs/software/appora10/linEM64T/en/inde
x.ht
m
Bryan Thomas
Senior Performance Consultant
Performance Tuning Corp.
http://www.perftuning.com
Email: bthomas_at_perftuning.com
-----Original Message-----
From: oracle-l-bounce_at_freelists.org
[mailto:oracle-l-bounce_at_freelists.org]
On Behalf Of Luca Canali
Sent: Monday, February 20, 2006 2:53 PM
To: oracle-l_at_freelists.org
Subject: RAC 10g interconnect resilient configuration
Hi,
I am trying to find a good HA configuration for the interconnect network
of RAC on Linux (10gR2 on RHEL 3). I have redundant switches and NICs
for the interconnect. From my tests and from Oracle documentation I see
that RAC can handle network (switch) failures, by simply bypassing the
failed network, while unfortunately 10gR2 clusterware can only be
configured to use 1 network (if that fails all RAC nodes but one go
down).
From metalink Note:220970.1 I found a reference to using NIC bonding
(unfortunately the link to the doc is broken, which is not a good start
for an HA doc).
I wonder whether implementing bonding is a good idea for the RAC
interconnect, besides the extra complexity, I am afraid of loosing
scalability regarding RAC (cache fusion) being able to load balance over
the NICs (as it does in the 'normal' configuration).
Any thoughts/ experiences to share on this?
Thanks,
L.
--
http://www.freelists.org/webpage/oracle-l
Received on Mon Feb 20 2006 - 16:04:08 CST
 |
 |