cURL / Mailing Lists / curl-library / Single Mail

curl-library

Re: low performance uploading files

From: Juan Ignacio Hervás <jihervas_at_iies.es>
Date: Thu, 27 Mar 2003 21:30:51 +0100

> There's a patch in there you can try (also available in the most recent
> 7.10.4 pre-release). This is included in the upcoming 7.10.4 release.

Thanks Daniel. I have downloaded curl-7.10.4 pre-5 and now it works.

> It is weird that Rich's tests proved 16KB to be a good size...

May be, but after lots of tests changing buffer size, I have found that the
best size is 40960.
Also, I detected that my TCP window size is 8 KB by default and I suppose
it has something to do with that.
Has Curl any kind of auto tuning to modify the TCP window size in order to
set the best size depending on the bandwidth? (I don't know if this is
possible or not)

thanks again.

At 20:28 27/03/2003 +0100, you wrote:
>On Thu, 27 Mar 2003, Juan Ignacio Hervás wrote:
>
> > Thanks RBramante. I have been reading your long discussion about this issue
> > and I have tested with a buffer size of 16384: the results have been very
> > bad!! (80 KB/s)
>
>It is weird that Rich's tests proved 16KB to be a good size...
>
> > Now I have a different problem that I suppose is a Daniel issue!
>
>It is!
>
> > I will try to tell you what I have seen in the curl source code:
> > My code is trying to upload a file resuming a broken upload. The file
> > pointer I pass to the library with CURLOPT_READDATA is copied to the
> > variable data->set.in. Then, it is copied to conn->fread_in. That's fine.
> > But in url.c (line 2595):
> > conn = conn_temp; /* use this connection from now on */
>
>I think your story indicates that you've hit bug #706624:
>
>http://sourceforge.net/tracker/index.php?func=detail&aid=706624&group_id=976&atid=100976
>
>There's a patch in there you can try (also available in the most recent
>7.10.4 pre-release). This is included in the upcoming 7.10.4 release.
>
>If the patch doesn't fix your problem, then I an very interested in a small
>source code for a program that can repeat this problem.
>
>--
> Daniel Stenberg -- curl, cURL, Curl, CURL. Groks URLs.
>
>
>-------------------------------------------------------
>This SF.net email is sponsored by:
>The Definitive IT and Networking Event. Be There!
>NetWorld+Interop Las Vegas 2003 -- Register today!
>http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en
Received on 2003-03-27