see
KTK>
/Klavs
> My boss, just discovered CacheFlow.. it's a cool lean, mean fighting
> machine.. but so is squid I tell him..
Sure squid is :)
> he's ready to let the two be set up, so we can compare their
> price/performance..
>
> Here's my problem.. I have set up squid 2.5 as an accelerator
> box (with the
> rproxy.patch)..
>
> but here comes my problem, in comparison with the cacheflow box..
>
> the cacheflow box, can be set up to refresh object's in the
> cache initially
> every 20 secs.. and then
> it adaptively - based upon how often the object actually
> changes it sets
> another refresh-rate.. thus saving
> capacity on the backend server..
see refresh_pattern.
Something I'd do is like (completely, utterly untested)
refresh_pattern jsp 0 0% 0
refresh_pattern html? 0 100% 10080
refresh_pattern jpg 1 100% 1440
etc etc.
format: "refresh_pattern" <regexp> <forced-valid> <percentage> <max-valid>
In this case the regexps are simple enough. The first line forces all
dynamic
content to be always re-fetched each time. The second says "html is
considered
fresh unless expressely reloaded for at most its age, and in any case not
more than 1 week", the third says "jpegs are valid for at least a minute,
at
most their age, up to a day".
KTK>this requires the server to state some sort of refresh-time?
KTK>the smart thing about the CacheFlow is that it sets
KTK>the refresh-time diffrently for every object - much more
KTK>detailed than any regexp pattern I could set..
The algo is very clearly explained in squid.conf, and is WAY more
customizable
than anything CacheFlow has to offer. Only "disadvantage", the resolution
is
in minutes, not seconds.
KTK> I get that.. 1 min. delay at min. - is ok.
KTK> it's just that when you have many objects, I don't want to have to
refresh them
KTK> all every minute - only the ones, that are often changed.. that's what
cacheflow
KTK> gives me.. without having to change code on the origin-servers.
KTK> as I see it, this should save requests between cache and
origin-server.
> This is the setup I have to compete with.. on squid, I can only set an
> update interval - to min. 1 minute.. -
> and I have journalists who updates somethings.. often.. and
> somethings not
> so often.. and they ofcourse want the changes
> to go on the site ASAP.. this is where the adaptive refresh
> is good.. cause
> perhaps they have to wait up to 5 min.. the first time,
> they change a page, after a longer time.. but the next time
> they change
> something, the adaptive refresh has lowered the time-between-refresh
> on that object.. while maintaining the higher
> time-between-refresh on the
> rest of the objects..
> This does save a lot of capacity, as old - never updated pages,
> automatically gets refreshed less often..
Just Ctrl-R (Ctrl-F5 in IE) while looking at that page, and anybody can
force a refresh for anybody.
KTK> and I don't want that.. cause then pages would be refreshed very
often..
KTK> as we deliver news, many clients actually leave the page open, and
press
KTK> Ctrl+F5 in their browser to see if there are any new news..
KTK>
KTK> Is it possible to set which IP's can force a reload via Ctrl+F5?
-- /kinkieReceived on Wed Nov 07 2001 - 02:58:17 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:03:56 MST