Gary Palmer wrote:
> Hi
>
> I'm setting up a squid and I have a machine with 7 2 gig drives
> to use as the cache store (another 2 gig for root, etc). The
> machine will be running FreeBSD. I have a couple of questions:
>
> - What is the best version of squid for a production environment?
> (even just for testing)
We tried a recent squid-1.2beta (beta 11 from memory) at one site and it
was a disaster.. We nearly had people trying to cut our throats since
things like hotmail.com were broken and a dozen other sites including
searching at microsoft.com were also hanging. Even on a test 1.2beta14
squid that I use locally, it still hangs forever if I do a search at
microsoft. (I think.. I have not tried it for as long as a week or so, so
it may have been before the machine was updated to beta14 with patches).
It may well be problems with the servers in question not handling
persistant connections or whatever, but you know for sure who the
customers will blame...
> - What is the best way to handle the drives? Leave them as individual
> drives and just tell squid to use 7 2 gig subdirectories, or
> stripe the drive?
If you stripe it into a super disk without redundancy, if you loose a disk
then you loose the entire cache. On the other hand, with squid 1.1, there
is a single table-of-contents, and I seem to remember squid not
particularly appreciating it if it lost one "chunk" of the cache and it
would return zero length replies for cache hits. However, the machine that
had that particular failure where a chunk of the cache was lost was also
having hardware (CPU overheating) problems, so it may not be entirely
squid's fault.
(Reminder: check those CPU cooling fans regularly! It's amazing what kind
of "impossible" processor faults turn up when the cpu is frying. The power
supply fan had failed too, the internal case temperature was around 85
degrees celcius (185 Farenheight).)
BTW; has anybody seen a source for fans (cpu and case) with tachometer
sensors for use with the new-fangled motherboards that can monitor them?
(preferably in .au)...
> Thanks,
>
> Gary Palmer
> NOC Supervisor
> Erols Internet
>
Cheers,
-Peter
-- Peter Wemm <peter@netplex.com.au> Netplex ConsultingReceived on Sun Feb 15 1998 - 02:42:43 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:38:53 MST