On Tue, 8 May 2007, Nicolas REMIA wrote:

> Im using libcurl in that way: i have 20 threads, and each thread manages an
> easy curl handle. Then, threads fight to obtain an url (in a shared pool) to
> download. Most of my requests are ranged requests.

But you do try to make the same handles transfer from the same host again,
right? I mean so that you can use persistent connections as far as possible.

> Should i manage myself to ask for several objects in one request using a
> ranged request like => 0-42,142-242,330-369. It means a local pool for each
> thread, waiting for a minimum number of files to download. Or should i use
> http pipelining ? My main problem is to dodge the latency, so make the less
> number of requests.

I would assume that you can do both and gain performance for both steps.

-- 
  Commercial curl and libcurl Technical Support: http://haxx.se/curl.html