Re: [squid-users] deep understanding of squid smp with disker and worker

From: Alex Rousskov <rousskov_at_measurement-factory.com>
Date: Thu, 13 Feb 2014 12:32:42 -0700

On 02/13/2014 11:54 AM, Dr.x wrote:
> hi all ,
> assume have
> 4 workers ( i used worker 4 in squid.conf)
> 2 disker ( i used two rock dir )
> 1 coordinator
>
> assume that google website has been cached on hardisks (disk1 and disk2
> which are represent by diskers process)
>
>
> assume that that userx is pumped to squid and requesting google.com
> now ,
> who will answer the user ??
> the worker ? or the disker ?

The worker. Diskers do not speak HTTP.

> well i will answer this question and need make sure that my answer is
> correct .
>
> as i understand , the responsible of answer the request are the workrers not
> the disker , the worker will try to solve it by memory , if it failed it
> will go to diskers ( responcilbe of disk caching and rock ) and try to
> server it from hardisks.

Yes.

> it will go to one of disker and if not found in one of them it will go to
> the 2nd disker and so on.

The worker does not exactly iterate diskers in search for a disk cache
hit. Instead, the worker uses disk indexes to check whether a disk cache
has the requested object. Those indexes are stored in shared memory so
the check is relatively fast. If the worker detects a hit, it sends the
corresponding read request(s) to the disker responsible for the
disk/index where the hit was found.

> if not found on memory or disker , squid will try to cache it if its size
> less or equal 32 kbit !

Approximately 32 KBytes (not 32 Kbits), provided you are using small
rock and the response is cachable based on HTTP rules and other factors.

HTH,

Alex.
Received on Thu Feb 13 2014 - 19:32:49 MST

This archive was generated by hypermail 2.2.0 : Fri Feb 14 2014 - 12:00:04 MST