[p2p-hackers] Testing scalability on a P2P network

[Ian Clarke]

On 1 Nov 2005, at 09:29, Frank Moore wrote:
> It may help, if this is what other people do in a similar situation.
> Greg mentioned testing it in the wild, but I'm not sure that's  
> practical for us.
> I know Bram Cohen did it for BitTorrent, but our dynamics are  
> slightly different.
> We have a p2p streaming app and it's going to be difficult to find  
> something that enough
> people are going to want to listen to or watch that will create a  
> network of a million users or more,
> which is the kind of number that we want to test against.
> If it was that easy to do, I guess we'd all be in the content  
> delivery business ;-)

Unfortunately testing it in the wild is the only way to be sure as it  
is virtually impossible to have a 100% accurate simulation of the  
Internet and user behaviour.  While this is true to some extent of  
most software, it is particularly true of P2P applications.

The next best thing to actual beta deployment is probably to use some  
kind of WAN simulator (try Googling for it) like "The Cloud".

I typically use an iterative approach to simulation and testing,  
starting with a very basic simulation of the network, which may not  
even take into account things like network latency, and then  
increasing the realism of the simulation in progressive stages,  
culminating in a beta deployment of the system.

Even if your beta deployment doesn't reach a large number of users,  
you can at least use it to verify the accuracy of your simulations  
for a smaller number of users.

The number of stages you want to employ depends on the time and  
resources available to your project.

Ian.