AW: AW: [patch] load images delayed

[Tobias Anton]

> Of course, if patch is very intrusive and makes
> other things harder shouldn't be included. 
As you can see, it is everything but intrusive, except that it breaks the
setAutoloadImages() facilities completely. What I'd like to know is: Will
a patch like this be accepted for CVS if it integrates smoothly into the
setAutoloadImages()-stuff of KHTML?
If not, would a patch be acceptable that expands setAutoloadImages() to
accept one of the values (no, yes, delayed)?

> It doesn't have to be precise. 

Precision is not my issue - reliability is. Imagine e.g. 100 users
browsing the web with konqueror. They share the bandwidth of an internet
connection of, say, 1Mbps. In that case, the available bandwidth per user
varies between 1Mbps and 10Kbps with rapid changes. It would be totally
pointless to try and estimate the available bandwidth from statistics of
preceding HTTP requests. That's why I wouldn't approach an adaptive
solution. A static switch, e.g. a config option, would be perfectly
sufficient AFAIAC.

> While I am thinking about it - why cannot
it be default behavior on *all* systems?

Good point. In fact, the optimal strategy does not only depend on the
bandwidth of the network connection, but also on its latency. If latency
is large compared to the bandwidth, one simultaneous connection may not
suffice to use the bandwidth optimally. Due to TCP handshaking and the
fact that we don't support HTTP pipelining, there are many situations in
which one machine waits for a specific message from the other.

As of now, we fill these holes by establishing multiple simultaneous
connections to one host, but then, the job of the main HTML page must
share the available bandwidth with the second connection, over which a
large image might be served.

This patch would avoid the described situation, but it would leave parts
of the network bandwidth unused. I doubt that this is the preferred
behaviour for the average user who I suppose to have a low-latency
high-bandwidth connection.

OTOH, with pipelining and one simultaneous connection per host, KHTML
would automatically behave as if it had this patch applied.

Tobias
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 1999 bytes
Desc: not available
URL: <https://mail.kde.org/mailman/private/kfm-devel/attachments/20060227/3a9c1819/attachment.bin>