Dmitry Suhodoev wrote:

hi, squid-users!

i have the squid 2.4-STABLE6 on my proxy server, and i very want reduce
consuption of traffic.

how i configure squid to release the following algorithm?

1. user try to download something (let it http://www.some-server.com/test.html)
from proxy (my squid).
2. squid must see to his own cache to hear size, date and time of cached test.html.
3. now squid looks to real http://www.some-server.com/test.html to hear his
size, date and time.
4. if first and second size, date or time different, only then squid will
download this real file, store in his cache and transfer it to user.
5. of course if file not exist in cache, then squid immediately downloads the
file.


i.e. i want make to my squid will ignore all "no-cache" and other http
attributes, and will use only this algorithm.

it is possible? if not, how i can reduce consuption of traffic by tuning squid
with action like this algorithm?


ps. or i'm full stupid, that thing this algorithm?


You're not full stupid, Dmitry, but probably at least half stupid. ;-)

If you have a standard proxy (not reverse proxy where you control the backend data), you /never/ want to disable all of the above safeguards against serving stale data in Squid. There are many circumstances in which the wrong data, such as data intended for a different recipient, could be served by the origin server. Expiry and cache-control headers are used in weird ways out in the real world, and it is hard to guess what your cache is going to run into. Ignoring some of them will cause problems with some sites. Ignoring all of them will cause problems with a /lot/ of sites.

You can tune some parameters to improve hit ratios, but completely disabling all of the checks isn't a wise choice in a cache with more than one user (ever, not just simultaneously).

What I would recommend, is to first turn up the percent in refresh pattern '.' from 20% to, say, 35%. Also increase the maximum lifetime from 4320 to something like twice that. Increasing the minimum age is also probably minimally risky...just don't go overboard--more than a couple of hours is probably asking for trouble.

That will probably gain you another 5% or so hit ratio. To get much beyond that, you'll have to start breaking the HTTP rules, which is not recommended. The least harmful might be to enable 'reload-into-ims'. Beyond that I wouldn't suggest use of any of the other refresh pattern options.
--
Joe Cooper <[EMAIL PROTECTED]>
Web caching appliances and support.
http://www.swelltech.com





Reply via email to