Re: Features/SmpScale

Sachin Malave wrote:
> On Mon, Dec 21, 2009 at 6:21 PM, Amos Jeffries <squid3_at_treenet.co.nz> wrote:
>> Sachin Malave wrote:
>>> Dear Amos,
>>> Thank you for updating wiki.squid-cache.org/Features/SmpScale, now
>>> ideas are more clear, I have already started working on the
>>> architecture that is proposed...
>> Thank you for your thanks.
>>
>> Which part and which layer are you working towards now?
>> (the earlier work you have already done was towards mid-layer)
>>
>> If it was top-layer there are a few speed bumps we've already identified and
>> not yet documented:
>>
>
> YES, IT IS.
>
>> Missing Pieces (most to least difficult):
>> * Mechanism to replace SquidConf:cache_dir (proposal cache_disk /media/path
>> SIZE ) - a set of disk entries. Master instance to load and portion off each
>> child instance with one or more disks. The child instance to determine what
>> types of storage (old SquidConf:cache_dir types) to be stored there and what
>> sizes will fit in the disk max given.come
>>
>> * Mechanism to handle many SquidConf:http_port and other listening port
>> directives - Master instance doing accept and passing to children? or all
>> children listening on all ports on a first-accept-wins basis?
>>
>> * Mechanism needed to replace existing method of master instance
>> "monitoring" for a dead child instance. Perhapse listening socket based?
>>
>> * Mechanism to split cache_mem between all child instances. Or do we leave
>> this as-is and call it a per-instance allocation?
>>
>> * Mechanism to keep unique_hostname unique per instance? do we even need
>> to? is it safer to consider the whole bunch of instances one unique "host"
>> and leave the existing code preventing complicated loops paths between
>> instances?)
>>
>> Amos
>
> I THINK YOU HAVE GIVEN ME ENOUGH MATERIAL FOR THIS ACADEMIC SEMESTER :)
> Before everything i will concentrate on all above mentioned points...
> I need some time for analysis..... will be back soon with solutions

:) hehe. Just pick one that appears easy. Or you may find something else
entirely we have not thought of. Any one of those bumps may take a
semester or several to get done and working. It will be around a year or
two before the top SMP layer is fully working.

Amos

-- 
Please be using
   Current Stable Squid 2.7.STABLE7 or 3.0.STABLE20
   Current Beta Squid 3.1.0.15