Re: Squid 2.2.STABLE2 & choice of parent

Henrik Nordstrom wrote:
>
> John Line (as web server manager) wrote:
>
> > Thank you! That seems to work - I see a few TIMEOUT_FIRST_UP_PARENT
>
> > 1437 FIRST_PARENT_MISS/cam1.sites.wwwcache.ja.net
> > 275 FIRST_PARENT_MISS/cam2.sites.wwwcache.ja.net
> ...
> > 41 TIMEOUT_FIRST_PARENT_MISS/cam0.sites.wwwcache.ja.net
> > 77 TIMEOUT_FIRST_PARENT_MISS/cam1.sites.wwwcache.ja.net
>
> I wonder... The distribution of TIMEOUT_FIRST_PARENT_MISS seems a bit
> disturbing I would say (cam0 selected "a lot", while not selected at all
> when there was no timeout). Or does this include the startup period?

No, it was a while after startup. However, cam0.sites was continually
reported as "TCP connection failed". It was also reporting ICP_MISS_NOFETCH
when it managed to send an ICP response (though until I got the debug
options sorted out, it wasn't clear which peer was sending those response).
So ... for requests done directly in response to ICP queries, cam0.sites
would not get used, but when using RTT estimates without having any ICP
responses, it looks as good as any of the others... It seems to handle HTTP
requests OK, in spite of the ICP_MISS_NOFETCH reesponses.

[I don't know what's wrong with that system ... I hadn't noticed until I was
looking at cache.log on the live server for comparison with 2.2, but it
appears to have been giving a lot of TCP connection failed messages for
several weeks (whereas the other parents have logged that sort of problem
only occasionally, probably during network maintenance).]

                                John

-- 
University of Cambridge WWW manager account (usually John Line)
Send general WWW-related enquiries to webmaster@ucs.cam.ac.uk