35110 - RFE: cache unchanging images longer

Status: RESOLVED INVALID

(Core :: Networking: Cache, enhancement, P3)

Description

[Brian Staples]

Here is an idea that would make people happier and have a faster browsing
experience. Mozilla should figure out which websites are visited most frequently
and download the graphics that always appear on the page like the title and
things it also could download graphics that don't always appear but that are
used commonly. Mozilla easily has the fastest rendering capabilities around but
this would make it even faster thus making the end-user happy. Just thought you
would like this idea I came up with.

Comment 1

[Jesse Ruderman]

Instead, how about keeping frequently-reloaded (but unchanging) images in the 
cache longer than images that are only loaded once or change each time they are 
loaded?  The browser would still check with the server to see if a new version 
is available in this case.

Comment 2

[Brian Staples]

Also it should download a new version of the graphic if a new version is
availible when it checks unlike NetSonic and Internet Explorer. It shouldn't
prompt the user in this case either, it just does its job.

Comment 3

[Gervase Markham [:gerv]]

Moving to Networking:Cache.

Gerv

Comment 4

[Jesse Ruderman]

Brista, it looks like this bug is changing from download images ahead of time 
to making the cache prefer larger images that are known not to change often, at 
the expense of possibly newer images that are smaller or change more.  Does 
that match your intentions, or should the cache-weighting idea be discussed 
elsewhere?

By the way, I have no idea how mozilla handles caching right now.  Maybe it 
already does that.

Comment 5

[Jesse Ruderman]

Changing summary to reflect e-mail discussion.  We want mozilla to keep images 
longer if they keep getting shown but never change their timestamp.  (Servers 
generally don't re-send images and pages if the browser-cached version's 
timestamp matches the server's timestamp for the object.)

An additional possibility I was thinking of was keeping large, unchanging 
images longer than small, unchanging images because those are the ones that 
take the longest to download.  Those take a lot more disk space, though, so 
maybe it should just be left at "consider how often the image changes in 
addition to when it was last used when cleaning the cache".

Should this apply to webpages as well?

Would someone familiar with mozilla's current caching behavior mind taking a 
minute to enlighten us as to what mozilla does now? :)

Comment 6

[Gervase Markham [:gerv]]

RFE, so Confirming.

Gerv

Comment 7

[davidm]

Heres the comment that describes our replacement policy. I would guess that we 
are already pretty much doing what has been suggested.

/**
 * Estimate the profit that would be lost if the given cache entry was evicted
 * from the cache.  Profit is defined as the future expected download delay per
 * byte of cached content.  The profit computation is made based on projected
 * frequency of access, prior download performance and a heuristic staleness
 * criteria.  The technique used is a variation of that described in the
 * following paper:
 *
 *    "A Case for Delay-Conscious Caching of Web Documents"
 *    http://www.bell-labs.com/user/rvingral/www97.html
 *
 * Briefly, expected profit is:
 *
 *   (projected frequency of access) * (download time per byte) * (probability 
freshness)
 */

Comment 8

[davidm]

any reason why this should not be marked invalid?

Comment 9

[davidm]

marking invalid since the caching policy already does this.