Network glitches costing 15 minutes a pop

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Robert Collins wrote:
> On Fri, 2008-08-15 at 11:19 +1000, Mark Hammond wrote:
> 
> 
>> Me too - but who is setting that timeout? :)
> 
> http://msdn.microsoft.com/en-us/library/ms819737.aspx documents TCP's
> behaviour on windows.
> 
> One possibility, simple retransmission delay would give us:
>  3 seconds (default RTT estimate) * 5 attempts = 15 seconds.
> 
> but 900 seconds is is 60 times as long, and it would require 5 dropped
> packets in a row.
> 
> Another possibility is high packet loss leading to many retransmits and
> finally 5 failures in a row.
> 
> I think a packet dump would be useful at this point.
> 
> -Rob
> 
> 

Just to give a counterpoint. I'm on windows here and haven't ever seen
this behavior. I don't use pycurl, though.

I don't know that it is pycurl, but we've had other issues (the
select/poll thing on linux.)

It certainly could be something to do with the latency from Sydney => LP.

John
=:->

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkik6WkACgkQJdeBCYSNAAMpuACgsHxw7+l68mkFbI12ufu9udG0
wlUAoK/VY9x8VRCahp1Hsxkzt6xbjL+u
=KlWu
-----END PGP SIGNATURE-----