On 2005-11-23T12:42:04, James Pan <[EMAIL PROTECTED]> wrote:
> The pink in the diagram shows the total taffic that captured by
> tcpdump, and the blue one shows the packets sent by hadev5 only.
> I do this to give meself a basic concept about how big the traffic
> overhead is. And now I'd say the traffic overhead is really small,
> especially when the cluster is in stable state.
Right, I once did this for 4 and 8 node clusters and the results were
similar (since Andrew fixed the election months ago ;-). Since the CIB
(which is the largest chunk of data we transmit) is only broadcast once
to all nodes, scaling is about linear for the traffic.
This is good to have and maybe we should capture the traffic of CTS runs
regularly so that we can make sure that no corner cases in our network
traffic go unnoticed. (It'd also help with debugging really obscure CTS
bugs ;-)
BTW, I grabbed the data with tcpdump and then used ethereal to visualize
this, which was quite easy and useful.
Sincerely,
Lars Marowsky-Brée <[EMAIL PROTECTED]>
--
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
