On Thursday 25 of February 2010 02:09:31 Yang Tse wrote:
> From that bug report it seems that in order to make that test pass on
> the build machine it was necessary to apply the patch that I'm copying
> below...

You're right, it made the particular buildhost happy, nevertheless it
doesn't sound like an enterprise solution to me :-)

> Current 7 seconds default was the smallest value that allowed this
> test to pass on all of libcurl's daily builds. There is no more magic
> behind this value.

It's simply machine-dependent, there is no such thing like optimal value.

> If increasing it to 8 or 9 makes it also pass on RedHat's build farm I
> see no problem in making the required value the new default for test
> case 1112.

I tried 10s and it was still failing occasionally. With 16s it seemed
to work fine in a loop.

> OTOH, if the test does not pass until a 16 second timeout is specified
> then something fishy, most probably related with the SSL handshake, is
> going on and should probably be further investigated.

SSL handshake is based on asymmetrical cipher, which is computational
expensive. I didn't consider there anything fishy since whole build
takes bigger amount of time on that machine.

> Could you verify which is the minimum timeout that makes the test pass
> on that busy system?

From my observation - 10s is not enough, 16s is enough.

Kamil
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2010-02-25