[erlang-questions] Cowboy vs Misultin

[Loïc Hoguin]

Hi Roberto,

On 05/30/2011 09:58 AM, Roberto Ostinelli wrote:
> 2011/5/30 Andy W. Song <wsongcn@REDACTED <mailto:wsongcn@REDACTED>>g
> it by "for ((x=2; x<50; x++)); do ip addr add 30.0.1.$x/24 dev eth0; done"
> 
> 
>     My machine info(Both machines are identical VMware VMs):
>     Debian GNU/Linux 6.0.1 (squeeze)
>     2.6.32-5-amd64
>     Erlang R14A (erts-5.8).
> 
> 
> Hi Andy, before I go further on this, are you really using R14*A* in
> your tests? A = beta. One of my client had serious memory leaks when
> using ejabberd over R14A. All that was required to solve the issue was
> to upgrade Erlang to a stable release and the memory leaks were gone.

Irregardless of this, AFAIK you are using lists for headers and
websocket data. This test doesn't send huge websocket data so it's
probably the headers of all the open connections, kept in your websocket
loops, that take a lot more space than for cowboy.

-- 
Loïc Hoguin
Dev:Extend