RE: rac network question

I have worked on 4 node RAC systems with this set up (against my advice), and it does work however when the network load goes up and crs can't communicate with the other nodes within an acceptable timeframe crs starts rebooting nodes. Even 50% network utilization for a few seconds can cause the reboots, so I really recommend against it. Also the network traffic doesn't have to be to or from the db tier, any traffic that saturates the backbone long enough to delay packets getting to/from the nodes can contribute to this. If you application sends a lot of data over the interconnect as did ours as it wasn't explicitly designed for RAC made it very susceptible. If you must go down this road, Make sure your sql is super tunes, and that you don't have any missing indexes.  

Ken Naim      

From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Andrew Kerber
Sent: Friday, January 11, 2008 9:49 AM
To: ganstadba_at_hotmail.com
Cc: Matthew Zito; oracle-l_at_freelists.org Subject: Re: rac network question  

I'm not even sure it will work. The private network is supposed to be for node-node communication. A lot of two node racs use a crossover cable for that connection, just to make sure nothing else will interfere. If this is truly a high visibility, mission critical database, this is simply a poor design.

On Jan 11, 2008 8:46 AM, Michael McMullen <ganstadba_at_hotmail.com> wrote:

I agree it will work, but isn't the private and public supposed to be physically separate, not logically?  

-----Original Message-----
From: Matthew Zito [mailto:mzito_at_gridapp.com] Sent: January 10, 2008 5:08 PM
To: ganstadba_at_hotmail.com; oracle-l_at_freelists.org Subject: RE: rac network question  

Actually, just so's we're all clear, with the VLAN support that the gentleman described originally, the interfaces will appear separate - eth0.1 and eth0.2 (note: different than eth0:1 and eth0:2). The traffic will be shared, but as long as the bonding works as it should, it just means that if a card is lost, both the interconnect and the VIP will fail over to the other link. IMHO, while this is suboptimal, it should work fine.  

Matt  

-- 
Andrew W. Kerber

'If at first you dont succeed, dont take up skydiving.' 



--
http://www.freelists.org/webpage/oracle-l