cURL / Mailing Lists / curl-library / Single Mail

curl-library

Re: CURLOPT_BUFFERSIZE and performance

From: Daniel Stenberg <daniel_at_haxx.se>
Date: Mon, 25 Oct 2010 14:59:58 +0200 (CEST)

On Mon, 25 Oct 2010, Luboš Doležel wrote:

> The whole {epoll - curl_multi_socket_action - write_callback} cycle has
> proved to be quite CPU intensive, half of the CPU usage is in 'system'.

Well, that info alone certainly isn't enough for anyone to tell if its normal
or not since it certainly depends on a lot of more data.

But I'll bite anyway: doesn't it struck you as weird? Why would the buffer
size be such a key value to this equation? A large size basically only reduces
the number of read() calls by some amount but surely that should in most cases
only be a matter of a very small percentage in difference.

It sounds like there's a problem somewhere.

-- 
  / daniel.haxx.se

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2010-10-25