[erlang-questions] large scale deployments and netsplits

[Bengt Tillman]

In the NETSim product we simulate the operation and maintenance traffic
of lots of different kinds of exchanges in  GSM or 3G-networks. We can
simulate thousands of exchanges and have to spread them over a large
number of Erlang nodes (up to a couple of hundreds) executing on a
number of computers (in the 10ths) in a local area network. 

We have had to set the net ticktime to 300 in order to keep the Erlang
nodes from losing contact with each other. The response times between
different Erlang nodes is not mission critical in our application - we
communicate between Erlang nodes in order to administrate the network of
exchanges and collect error reports. The mission critical traffic is
TCP/IP (or other O&M protocols) between a simulated exchange and the
system our customers want to test (OSS).

BR,
Bengt Tillman

> Ulf Wiger wrote
> 
> Joel Reymont wrote:
> > 
> > On Sep 14, 2009, at 4:48 PM, Ulf Wiger wrote:
> > 
> >> What is your setting for net_ticktime?
> > 
> > Is there a way to set net_ticktime programmatically, once 
> the node is 
> > up and running?
> 
> net_kernel:set_net_ticktime(TickTime [, TransitionPeriod])
> 
> The default tick timeout is 45 < T < 75, with TickTime being 
> 60 seconds.
> 
> See http://www.erlang.org/doc/man/kernel_app.html
> 
> Normally, and especially for systems where response time is 
> an issue, the default ticktime is too long - not too short.
> If you have network connections that are not responding 
> within at least 45 seconds, it seems as if you have some 
> rather severe overload on your links.
> 
> BR,
> Ulf W