[AusNOG] squid peering mesh

[Julien Goodwin]

On 01/02/11 17:46, Adrian Chadd wrote:
> On Tue, Feb 01, 2011, Julien Goodwin wrote:
> 
>> We run squid clusters of up to a few hundred megabits, and have mostly
>> stopped even bothering to peer with boxes in the same cluster (same
>> rack, same switch, etc). The latency hit just isn't worth it these days
>> given how little is cachable.
> 
> Which is weird, because the caches I setup and maintain still save 20-30%
> of HTTP bandwidth without too much hackery. One goes above 30% with some
> hackery.

That's still the case, but most of the benefit of peering is increasing
your effective cache size, and even with tens of thousands of users
hitting them 50GB of cache, trivially servable from a single host these
days satisfies the vast majority of requests.

So perhaps that was better written "... how little is cacheable in
comparison to standard server disk sizes"