Re: Network interconnect traffic on RAC


CRS messages are very small (around 100s of bytes) and CSS messages
are less than 1K.  Also you have around 30 seconds to evict the node
during any of these message-not-reaching-target-due-to-congestion
situations. Its very very rare case of node evictions due to

On the other hand, if you suspect you are flooding the interconnect,
you will have a bigger problem. Most common interconnects I see these
days are 10gigE.
10gigBits=1GigBytes/sec=1024MBytes/sec=1024*128*8K=128M*8K ~100M
blocks/sec. Unless all your data virtually lives in the interconnect
you will not saturate the pipe.

To answer your original question, AWR has a section called "Estimated
Interconnect Traffic" which is built on top of CR_Sever views. You can
query the gv$ CR server views to find out the number of blocks served
during that time and see how much bandwidth you may use.


On Wed, Feb 10, 2010 at 12:34 PM, Karl Arao <karlarao@xxxxxxxxx> wrote:
> Hi List,
> How do you measure the network traffic going on your RAC interconnect?
> on the Oracle and OS side...
> Can you share the efficient tools that you use?
> I'm trying to figure out (aside from the GC wait events popping out on
> the AWR)...  how would you know if your interconnect is congested?
> Let's say I had some RAC node evictions, I want to measure if it's
> really the interconnect that is congested (I'm looking for the
> numbers).. the same way that I'll be looking on the CPU scheduling and
> disk IO issues...
> - Karl Arao
> --

Other related posts: