On Fri, 30 Nov 2007, Enrico Stark wrote:

> For the 512 files the connection died 125 times, so in average every fourth
> transfer the connection was built up again.

I copied in a bunch more files in my test dir now and I just uploaded 91
(pretty small) files just fine, all using the same connection (on localhost).

> Unfortunately there is no pattern recognizable, since it is always the same,
> the only thing that differs is the filename and the file size.

And it isn't about data amounts or possibly time passed ?

What if you only upload 20 files, all being 3K big each?

Have you tried to use Wireshark or similar to try to detect what actually
happens on the network, I mean why libcurl thinks the connection is dead? I
mean, it could possibly _be_ dead! libcurl even handles the case pretty nicely
if I may say so...

> I'm using libssh2 version 0.18.

I use 0.19.0-CVS but the difference is really very slim and I doubt it
matters.

> The test is done with a remote machine, in the same network, no firewall or
> proxy is between.

I'll do some further testing soon and upload the files from my laptop to my
desktop over wifi to see if that makes things different here.

> * About to connect() to localhost port 22 (#0)
> * Trying ::1... * Failed to connect to ::1: Network is unreachable
> * Error 0
> * Trying 127.0.0.1... * connected

This happens because libcurl gets (at least) two addresses returned when it
resolves 'localhost' (one IPv6 address and one IPv4 address) and while you
seem to have ipv6 support you don't seem to have a ipv6 network...

> I'm not sure about the sshd version, the output that the sshd gives me is:
> OpenSSH_4.5p1, OpenSSL 0.9.8e 23 Feb 2007

Pretty close to what I have: OpenSSH_4.6p1 with OpenSSL 0.9.8g.

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