On Sat, Jan 23, 2010 at 09:07:10PM +0100, Johannes Hofmann wrote:
Rogut??s Sparnuotos :
I have skimmed over hundreds of photos on flickr.com with both, vanilla dillo-2.1.1 and dillo-hg+connlimit2, with proxy and without. Browsing seems to work fine with all variations (I didn't try to analyze the packets), but I had a hard time trying to see the difference: I guess my internet connection is too fast and my X server too slow. Since I have a good internet connection, the images loaded faster without the patch, but sometimes the ones at the top of a page appeared sooner when the patch was applied.
Openning all the connections at once helps speed by making the wait for connections parallel for all images. This is noticeable with a big network latency.
Yes, that's expected. The purpose of the patch is not to improve loading speed or user experience, but to play nice with servers and infrastructure. Modern sites especially big ones like flickr are well prepared to handle the load caused by the few dillo users out there. Nevertheless it's important to use network resources in a sensible and compliant way.
I expect most of the mainstream servers to queue multiple requests and thence serve from their queue as much as they want. Other applications may have trouble with a herd of requests at the same time and hence prefer a connection limit. Pages with hundreds of non-tiny images (e.g. some blogs) may *feel* better with throttled requests; here users see the nice effect of images loading "in order".
Since dillo and the servers seem to work with unlimited connections, perhaps defaulting to a limit of 6 wouldn't hurt?
Yes, the current limit is certainly debatable. We could even make it configurable. Any opinions on this?
Configurable with a dillorc option would be welcomed. We could comment suggested values for dialup, low band, broadband, ...
Dillo crashed twice in the beginning of my testing, but I am not sure anymore which build it was (because I've applied the gnutls-0.patch at first - https seemed to work). I couldn't reproduce the crashes afterwards.
Please keep an eye on potential crashes. Those need to be ironed out of course.
I remember when making the first connection limit patch to have apprehensions on the stop/abort process. This is very complex inside Dillo. If you want to crash-test, try going to pages then pressing stop (or quiting that tab), or opening a new one while the other loads, then aborting the first one. Things like that will make dillo stop, clean and keep going. That's the gist. Thanks for your feedback. -- Cheers Jorge.-