Re: Squid 2.2.STABLE2 & choice of parent [updated second patch]

A unpatched Squid 2.2.STABLE2 responded to this situation by going
direct if there was lost ICP packets to one or more of the peers (bug,
received replies from other systems should not be ignored only because
one was lost).

A related improvement has also been produced to make ICP parent peerings
less sensitive to ICP packet loss.

I think John Line was happy with the end result of these patches on his
system, but maybe he is a bit subjective due to the UK network charge
policy...

My opinion this improvements is a step in the right direction, but there
should be some noticeable priority penalties to peers with packet losses
(or even worse: TCP connection failures), so it can be done much better.

Duane has committed the first (the bug fix) to the main source tree.

And in the same discussion (but not related to this) was another
improvement made to how the dynamic timeout used in ICP peerings when
icp_query_timeout is unset. This was made with the assumption that most
siblings are close and parents usually remote, and that there is not
much use in peering with a sibling which is much slower than the average
parent.

/Henrik

Martin Hamilton wrote:
>
> -----BEGIN PGP SIGNED MESSAGE-----
>
> John Line writes:
>
> | The JANET national cache (or at least, two of the three systems there that
> | we use as parents) seem to be having even worse problems today - both the
> | systems at Manchester are now failing to accept TCP connections most of the
> | time and are responding with ICP_MISS_NOFETCH for misses. Useful for
> | checking the latest patch, though!
>
> We've been seeing packet loss of as much as 70% to/from this cache
> network and the rest of the world :-( Should be a good test of how
> Squid behaves when your local bit of the Internet goes into
> pathological failure mode!
>
> -----BEGIN PGP SIGNATURE-----
> Version: 2.6.3i
> Charset: noconv
>
> iQCVAwUBNzczfdZdpXZXTSjhAQESRgQAn/HeRCiuxvEAeiUJZihXEB2SjP/E85eH
> X97HaNxSiyBfWN/qHLHG2q2efre3IlB99jbihe8E+xdy7icYfRe6XIawfcIxNg56
> 0HV3IsM3NZS1eTlI8FxgucV2lkH+/x2ZXOuWhgkShDDLEsJM2UOfAqoBvRKl1Zw7
> ukH165Uij0o=
> =xnQG
> -----END PGP SIGNATURE-----
Received on Tue Jul 29 2003 - 13:15:58 MDT