Battsetseg.M wrote:
Hi all,
We’re medium sized ISP and want to implement caching.
After doing some research, we decided to use load balancer with 4-5 servers.
Our exiting gateway bandwidth is totally 300Mbps. After doing some
calculations included in O’Reilly ‘Web Caching ‘ we got to have a 2800GB of
cache-dir and a huge amount of memory. The question is did I miscalculate?
Or anyone has a better suggestion?
The calculation specifics are:
Bandwidth 300Mbps
Total 80% of the bandwidth is HTTP
Cache-hit: 40%
Cache-miss: 60%
Non-cacheable objects: 20%
Any help is appreciated.
What calculations? I don't think it quite works that way.
An estimated conservative HIT rate would be around 30% and it varies.
The non-cacheables are more likely to be the same as MISS rate. Or they
would become HITs themselves under peak loads.
But, if you have no caching now the HIT savings will be a net gain even
if its only 1% using a 10MB cache. Which brings the good news that any
cache size you can afford can be useful.
The approach I'd recommend is to spec out what you can afford and expand
it over time. That includes starting by only caching a small segment of
the client base to see how many a single squid (or pair for redundancy)
can handle it.
Amos
--
Please use Squid 2.7.STABLE4 or 3.0.STABLE10
