But how/when is the absence of a file evicted? It would seem like a very
small gain for the cache-system to throw out since it would take up very few
(0) bytes. Also, Jacek Piskozub, said in the bug that the caching is per
session. Is this intended/a bug/incorrect?

In either case, the second formula [1] can still be used, just use the
average time in the cache as X. Lets say that it takes 5 days for a cache
entry to be thrown out, then set X=5, any netwerk people has any good
numbers? If the caching should indeed be per session, then X could be set to
1.

For people testing this: Please use an "average" timeperiod for testing, it
would be really interesting to get some real numbers as basis for a
well-conditioned decision.

/ Jonas Sicking

[1] Hits = distinctUsersPerXDays * 0.3 * 30/X

"David Hyatt" <[EMAIL PROTECTED]> wrote in message
[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
> The icon is not cached "forever".  It simply has no specified
> expiration.  That just means it won't be doomed based only off some
> expiration date.  It can still be removed from the cache as the cache
> fills up and needs to evict items.
>
> dave
> ([EMAIL PROTECTED])
>
> Jonas Sicking wrote:
>
> > A lot of oppinions has been expressed with regard to if the favicon
should
> > be default on or off since it might spam webservers with requests to a
> > non-existing file.
> >
> > It would be really interesting to get some hard numbers on this. Just
> > looking at the current logs will not really say anything since very few
> > people browse with a mozilla with this pref turned on. So we need to
come up
> > with some way to approximate the number of 404s per (for example) month
in
> > the event of a browser with, say, 30% marketshare using the current
> > configuration.
> >
> > Since the absence of a /favicon.ico is cached the number of 404-ing
requests
> > will be much lower then the numbers of pagehits. Brendan says that the
> > absense is cached "persistently and with never-expire", does that mean
that
> > mozilla won't request /favicon.ico again unless the user manually clears
the
> > cache? In that case the number of 404s will be approximatly equal to the
> > number of "new users" every month * 30%.
> >
> > If it's not possible to extract the number of new users from the logs i
> > think that the number of new IP-addresses * 1.5 is a good enough
estimation.
> > There are probably more then 1.5 user per IP on average, but all users
> > probably don't visit the site. If someone have a better number then 1.5,
> > please speak up, my guess is very uneducated.
> >
> > However it seems a bit wrong to me that a resource is cached "forever".
What
> > if a site want to start supporting /favicon.ico? Will only new users see
the
> > new icon? IMHO a resource should be reloaded at least sometime so that
if
> > the resource appears/changes we will eventually catch it.
> >
> > So say that we reload every 2 weeks. That means every user will reload
> > /favicon.ico once every 14th day, which means that the number of 404s
will
> > be "number of destict users during 14-days" * 30% * 30/14.
> >
> > So, we've got:
> >
> > Hits = newUsersPerMonth * 0.3                 if we cache indefenetly
> >
> > Hits = distinctUsersPerXDays * 0.3 * 30/X     if we refetch every X days
> >
> > Where IP-addresses * 1.5 could approximate number of users. IMHO the
right
> > thing would be to use the second formula with X ~= 14.
> >
> > So it would be great if someone with access to the logs to a rather
heavily
> > used site could run these formulas and compare that to the number of
> > "normal" 404s.
> >
> > / Jonas Sicking
> >
> >
> >
>



Reply via email to