Henrik Nordstrom wrote:
> On Thu, 13 Oct 2005, [ISO-8859-1] Hendrik Voigtl�nder wrote:
>
>>> Should be more or less the same when the parents fails to resolve..
>>
>>
>> But in this case the child should fetch direct instead of displaying
>> the error page?
>
>
> That is the general intention of Squid request forwarding.
>
Hmm, I still dont get why the child displays the error page when the
parents fails to resolve. I think I will use the workaround with the
line monitoring.
>> Any opinion about establishing sibling relationship between the backends?
>
>
> sibling relations does not help the problem you are seeing, but may
> improve hit ratio.
>
Improving the hit ratio is why I want to do this. Are there any best
practices, e.g. icp?
> The same improvement in hit ratio for a lower overhead cost can be seen
> if you configure the child caches to use carp to select which parent to
> use. When using carp the childs always sends the same URL to the same
> parent (forwarding selection based on a hash of the requested URL).
I dont think this will improve hit ratios as the the parents are
non-caching ( to slow - ancient Compaq Proliant 1500's ) but this will
IMHO allow load-balancing without the drawback of using round-robin
parents.
I have seen trouble with some webpages as both parents have different
IP'd due to the DSL connection (session based stuff).
Am I right? Will carp stick to one parent depending on the whole url or
just the hostpart?
Regards, Hendrik Voigtl�nder
Received on Mon Oct 17 2005 - 03:37:15 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Nov 01 2005 - 12:00:04 MST