Search squid archive

Re: about the cache and CARP

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Aug 23, 2011 at 9:01 AM, Amos Jeffries <squid3@xxxxxxxxxxxxx> wrote:
> On 24/08/11 00:47, Carlos Manuel Trepeu Pupo wrote:
>>
>> 2011/8/23 Amos Jeffries<squid3@xxxxxxxxxxxxx>:
>>>
>>> On 23/08/11 21:37, Matus UHLAR - fantomas wrote:
>>>>
>>>> On 16.08.11 16:54, Carlos Manuel Trepeu Pupo wrote:
>>>>>
>>>>> I want to make Common Address Redundancy Protocol or CARP with two
>>>>> squid 3.0 STABLE10 that I have, but here I found this question:
>>>>
>>>> the CARP that squid supports is the "Cache Array Routing Protocol"
>>>> http://en.wikipedia.org/wiki/Cache_Array_Routing_Protocol
>>>>
>>>> - this is something different than "Common Address Redundancy Protocol"
>>>> http://en.wikipedia.org/wiki/Common_Address_Redundancy_Protocol
>>>
>>> Well, technically Squid supports both. Though we generally don't use the
>>> term CARP to talk about the OS addressing algorithms. HA, LVS or NonStop
>>> are
>>> usually mentioned directly.
>>
>> Thanks for the tips, from now I will be careful with the term.
>>
>>>
>>>>
>>>>> If the main Squid with 40 GB of cache shutdown for any reason, then
>>>>> the 2nd squid will start up but without any cache.
>>>>>
>>>>> There is any way to synchronize the both cache, so when this happen
>>>>> the 2nd one start with all the cache ?
>>>>
>>>> You would need something that would synchronize squid's caches,
>>>> otherwise it would eat two times the bandwidth.
>>>
>>> Seconded.
>>>
>>> If the second Squid is not running until the event the cache can be
>>> safely
>>> mirrored. Though that method will cause a slow DIRTY startup rather than
>>> a
>>> fast not-swap. On 40GB it could be very slow, and maybe worse than an
>>> empty
>>> cache.
>>>
>>> NP: the traffic spike from an empty cache decreases in exponential
>>> proportion to the hit ratio of the traffic. From a spike peak equal to
>>> the
>>> internal bandwidth rate.
>>>
>>> PS.  I have a feeling you might have some graphs to demonstrate that
>>> spike
>>> effect Carlos. Would you be able to share the images and numeric details?
>>> I'm looking for details to update the 2002 documentation.
>>
>> Thanks to everyone, you guys always helping me !! Now I have a few
>> problem with Debian and LVM, until I solve it I can't do it anything.
>> But here another idea:
>>
>> I put the two squid in cascade and the Master (HA) make the petitions
>> first to the second squid and if it down go directly to Internet. The
>> both squid will cache all the contents, so will be duplicate the
>> contents, but if someone go down, the other one will respond with all
>> the content cached.
>>
>> It look like this:
>>
>> client --->  Server1 --->  Server2 --->  Internet (server1 and server2
>> will cache all)
>> Server1 down
>> client --->  Server2 --->  Internet (server2 will cache all)
>> Server2 down
>> client --->  Server1 --->  Internet (server2 will cache all)
>>
>> What do you think ?
>
> Looks good.
>
> Check your cache_peer directives connect-fail-limit=N values. It affects
> whether and how much breakage a clients sees when Server2 goes down. If that
> option is available on your Server1 squid, you want it set relatively low,
> but not so low that random failures disconnect them.
>
> background-ping option is also useful for recovery once Server2 comes back
> up.
>
> Amos
> --
> Please be using
>  Current Stable Squid 2.7.STABLE9 or 3.1.14
>  Beta testers wanted for 3.2.0.10
>

Everything it's working fine !! Until now they are still in LAB mode,
but with excellent results in tests. Now I would like to improve the
mechanism to HA of my servers. Any other idea how to improve the work
that I made until now. (I just make squid with UCARP in Debian)



[Index of Archives]     [Linux Audio Users]     [Samba]     [Big List of Linux Books]     [Linux USB]     [Yosemite News]

  Powered by Linux