Re: [squid-users] it was a slow death

On 25/12/10 01:32, Optimum Wireless Services wrote:
> On Wed, 2010-12-22 at 16:54 +1300, Amos Jeffries wrote:
>> On 22/12/10 04:49, donovan jeffrey j wrote:
>>> Greetings
>>> i discovered the culprit to my woes as my internet connections slowly died. It was my 2 cache drives. As they would fill, and swap, and fill, and swap.. well you get the picture. Both drives just burned up and won't mount.
>>>
>>> So im running a cache_less system, which we are finding is really quick.
>>>
>>> does this look right for intercept only no cache ? are there any performance adjustments I can do ?
>>> squid 3.1.9
>>>
>>> http_port 10.0.1.1:3128 transparent
>>
>> not performance exactly but "transparent" should be written "intercept"
>> in 3.1+
>>
> Hi Amos.
>
> Sorry to hijack this thread.
>
> Are you saying that for version 3.1+ we should use:
>
> http_port 10.0.1.1:3128 intercept
>
> instead of transparent?

Yes. Exactly so.

>
> I'm currently using 3.1.9 like this:
>
> http_port 172.16.0.1:3128 transparent disable-pmtu-discovery=off
>
> Is that correct?

Same there.

The PMTU setting depends entirely on your local network situation.
Disabling is a last resort. Traffic as a whole will work much better if
the problem requiring it can be resolved. It's often just a matter of
finding and fixing an ICMP config somewhere (hopefully under ones own
control or a that of friendly external admin).

>
>> Otherwise it looks just fine.
>>
>> The other performance adjustmenst would be kernel and TCP stack things
>> to make ports available more frequently (avoiding some TIME_WAIT) and
>> accept jumbo packets etc. I'm not sure on the exact sysctl knobs to
>> tweak, they should be easy to find if you have not done them already.
>>
>> FWIW: 3.1.10 is now in the release process with several memory
>> consumption fixes.

Amos

-- 
Please be using
   Current Stable Squid 2.7.STABLE9 or 3.1.9
   Beta testers wanted for 3.2.0.3