On Sat, 30 Apr 2011, Ben Noordhuis wrote:

>> Any suggestions on what could cause this ?
>
> Maybe I'm not quite following you but isn't this to be expected?
>
> There is a fixed amount of CPU time to go around and it's being sliced
> thinner and thinner as the number of threads grows. How can the response
> times *not* go up?

It would make sense if the 10 threads used 100% CPU, but usually getting 10
simultaneous transfers don't consume all CPU so adding more threads should not
make them run (much) longer.

I don't understand the numbers, but I don't know Boost at all and I don't have
it installed so I can't repeat the experiment very easily in my end.

Radu, can you write up an example repeating the symthom using ordinary
pthreads so that we can try it out easier without extra dependencies?

-- 
  / daniel.haxx.se
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html