Re: [squid-users] Same cache_dir for many squid process

From: Amos Jeffries <squid3_at_treenet.co.nz>
Date: Sat, 01 Jun 2013 18:00:53 +1200

On 1/06/2013 2:59 p.m., csn233 wrote:
>> if you correctly configure your parents to not cache anything, then
>> you'll have no duplications. In that case, only the squid you'll choose will
>> manage cache on disk.
>>
>> it can be as simple as a
>>
>> cache_dir null
>>
>> to the parent squids.
> If the parents have cache_dir null, those parents would simply be
> relaying the requests to the origin servers.

No. Critical detail: there are no siblings in a chain.

That hierarchy page has two images at the top for a reason. To clarify
that there are two distinct relationship types of parent-child
(chaining) and sibling-sibling (parallel scaling). The properties and
behaviour of each relationsup and distinctly different. And like I wrote
in that top section, they can be combined into many different layouts to
perform many very different behaviours.

Leonardo has been talking about a "correctly configured chaining". Using
sibling relationship in a chain is *not* correctly configured, the
existence of sibling relationship converts a chain hierarchy into some
other type of hierarchy. A chain is simply a sequence (chain) of at
least 2 proxies like so: client<->A<->B<->C<->D<->origin with so side
branches or bypasses.

> They do not add any value
> - the child cache(s) can simply send the requests direct. If fact,
> you are adding overheads, the same 1 single request has to go through
> the child and then the parent, ie 2 separate processes to process the
> same 1 request.

CARP is the type of hierarchy you want for scalability if SMP is
unavailable. It is two-layer and each layer is independently scalable.
The gateway layer (childs) does not cache, but simply does access
controls on clients and determines which cache-layer (parents) proxies
is storing the URL being fetched to pass it through on the way to the
origin.

Note that CARP typically has no sibling relationships. Yet provides
cache de-duplication, failover, and scalability for high-performance
beyond what a single or several parallel Squid caches can offer. It does
require specific hardware and configuration arrangements however.

> With SMP, if you have 2 processes, you can process 2 separate requests
> simultaneously.

Amos
Received on Sat Jun 01 2013 - 06:01:01 MDT

This archive was generated by hypermail 2.2.0 : Sat Jun 01 2013 - 12:00:07 MDT