Re: [Openvpn-users] Re: Re: Scalability?

[James Yonan <jim@xxxxxxxxx>]

On Wed, 2 Mar 2005, Jamie Lokier wrote:

> James Yonan wrote:
> > I'm hoping to get pthread support back in for 2.1.  Having said that, I 
> > don't think it's going to make much of a difference in most cases.  Where 
> > it will matter is if you have scripts which take more than 100 
> > milliseconds to execute.
> 
> Does lack of pthread mean that new connection, rekeying calculations,
> and scripts, pause the tunnel while they happen?

New connections don't stall the tunnel.

Scripts do stall the tunnel.

Rekeying calculations stall the tunnel for a short interval (~40 
milliseconds for 1024 bit RSA keys on a 2Ghz P4).

The OpenVPN code itself has very good latency numbers.  The problem is 
when other scripts or libraries are called.

I'm currently exploring two possible ways to implement multithreading:

(1) Run a two-thread model where the bulk of OpenVPN runs in the main 
thread, but high latency operations are offloaded to a worker thread.

(2) Run an n-thread model where all threads are simultaneously processing 
the main event loop.

The main issue here is development time.  I estimate it will take 1 to 2
months of work to produce an initial beta.  That seems like a lot of time
to spend on a feature most people will never even notice.  I've seen 
people running VoIP and even videoconferencing over the current 
single-threaded code and never even noticed any latency issues.

I'd be interested to know if people care enough about multithreading to
help fund its development.

James

____________________________________________
Openvpn-users mailing list
Openvpn-users@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/openvpn-users