RE: RAC Infiniband Questions

From: Carlson,Todd <Todd.Carlson_at_wwt.com>
Date: Wed, 1 Sep 2010 09:56:58 -0500
Message-ID: <9013C1CFFF300F4CAB31B78DC7669E010D17365296_at_PRODCMS1.wwt.local>



Hey Matt,

Thanks for the response! Below is the output from DEV. The problem that we are having is that we have extensive experience with IPMP, but no experience with RHEL binding. As a result, when we bound 2 channels across cards, we really didn't know what we were doing and we did get the cluster to install, but then the traffic across the interconnect stop until we bounced the switches. Weird.

Do you know of a step by step process to bind the channels for RAC that we could see? Also, since we have 2 cards with 2 channels, we would end up with two bound ports for the interconnect. When we do the cluster install, would we select both for the private interconnect?

/u01/grid/11.2.0/bin>oifcfg iflist

eth0  10.2.9.0
eth1  10.2.9.0
ib0  192.168.0.0
ib1  192.168.0.0
ib2  192.168.0.0
ib3  192.168.0.0

/u01/grid/11.2.0/bin>oifcfg getif

eth0 10.2.9.0 global public
ib0 192.168.0.0 global cluster_interconnect

Todd

From: Matthew Zito [mailto:mzito_at_gridapp.com] Sent: Tuesday, August 31, 2010 3:37 PM
To: Carlson,Todd; oracle-l_at_freelists.org Subject: RE: RAC Infiniband Questions

You use the bonding driver - just like you do for Ethernet, iirc. What problem are you experiencing? What do your ifcfg- files look like?

Matt



From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Carlson,Todd Sent: Tuesday, August 31, 2010 4:06 PM
To: oracle-l_at_freelists.org
Subject: RAC Infiniband Questions

Hey Guys,

We are building out our RAC environment on 11.2.0.1.2 with RHEL 5.5 on Sun x4270 servers. We are running into problems trying to bind our Infiniband connections together. In SAND & DEV, we have 2 node clusters. Each node has 2 HCA's (Sun Dual Port 40Gb/sec 4x Infiniband QDR Host Channel Adapter), 1 & 2 with 2 ports, A & B. We have 2 Sun 36 port Infiniband switches. What we are trying to do is bind 1.A with 2.A and bind 1.B with 2.B. However, there is very little documentation from Sun/Oracle or RedHat on how to do this and all of our attempts have failed. So, right now these environments are using one connection between the nodes using UDP.

We are currently building out TEST and we need to have this working by the 10th of Sept. We will then rebuild DEV & SAND to have the binding working correctly. So, I am hoping/praying that you can guide us here on how we go about configuring the binding of the IB ports. Is there some documentation that you would send me? We have read "RAC Support for RDS Over Infiniband [ID 751343.1]" and are essentially stuck at step #2.

On a Similar vein, once we get the channels bound, we would then have 2 private, bound interfaces to use. In the cluster install (step 6 of 16), would we select both of them or just one?

Thanks for your help here, I really appreciate it!

Todd Carlson
Manager - DBA/ERP & EUC Teams
World Wide Technology
(314) 301-2788

--
http://www.freelists.org/webpage/oracle-l
Received on Wed Sep 01 2010 - 09:56:58 CDT

Original text of this message