[squid-users] squid with SMP registeration time out when i use 10K opened sessions

[Ahmad Alzaeem]

Hi Amos , thanks for reply 

Regarding to your description

If I have 10K ips with 10k listening ports ......

Will each squid process handle 10 K ?
Or I need to distribute the ips/ports to each process ???

cheers

-----Original Message-----
From: Amos Jeffries [mailto:squid3 at treenet.co.nz] 
Sent: Friday, September 25, 2015 12:44 AM
To: Ahmad Alzaeem
Subject: Re: [squid-users] squid with SMP registeration time out when i use 10K opened sessions

On 25/09/2015 9:27 a.m., Ahmad Alzaeem wrote:
> Hi amos
> I have alredy 5.2

3.5.9 is the latest security release.

> 
> All squid traffic go to one of cpus and it load it 100 % I don’t care 
> caching or other thing
> 
> All I need is to load like 20K ports and be balanced to the cores

You wont get balance on the cores. They actually work *better* when unbalanced. It makes the core L2/L3 RAM cache work with closer to optimal contents.

With 3.5.7 or later run several Squid with:

 squid -n squid1
 squid -n squid2
 squid -n squid3
 ...

The "${service_name}" variable in suqid.conf will become that "squid1", squid2" "squid3", etc

You need to make your squid.conf contain:
 include /etc/squid/${service_name}-ports.conf
 pid_filename /var/run/squid/${service_name}.pid
 cache_log /var/log/squid/${service_name}-cache.log

... and so on for the other directives listed in <http://wiki.squid-cache.org/MultipleInstances>

Make the squid1-ports.conf etc config files listing your http_port's and a unique_hostname for each Squid instance. Maybe access_log as well.

Also set cpu_affinity for each instance to tie them to different cores.

Amos