- From: <juan@canonicalscience.com>
- Date: Mon, 14 Apr 2008 06:00:48 -0700 (PDT)
- To: <public-html@w3.org>, <www-math@w3.org>
- Cc: <davidc@nag.co.uk>
David Carlisle <davidc@nag.co.uk> wrote: > It took me longer to copy that URI to the browser than it took FF to > render it on my machine, perhaps you have some local setup problem > that is colouring your view. This rendering performance has been verified on different hardware over both Windows and GNU/linux platforms for both default and non-default configurations of FF 1.0, 1.5, and 2.0. Existence of performance problems is also noticed by Mozilla http://www.mozilla.org/projects/mathml/authoring.html {BLOCKQUOTE Avoid extraneous tags, and/or tags with content that collapse into nothingness [...] it has a number of flaws in terms of performance and memory consumption: * Overhead to fetch the file on the net * All the superfluous <mrow> are kept in the DOM * CSS frames are created for each and every superfluous <mrow> * Time is wasted to reflow (format) them. [...] Avoid ignorable whitespace at the beginning and the end of token elements [...] MathML is already verbose, and you don't want to add to that verbosity. Do you? Note also that, in the code, price is paid in the loops used to trim and collapse whitespace. Remember, on the web, the same persons that can wait for the termination of a lengthy TeX run from the command line seem tempted to go elsewhere when a site is slow. By keeping extra whitespace out of your markup, you are skipping out of several loops in the code, thus enabling a speedy and better browsing experience to visitors of your website. } Performance issues when rendering MathML on Mozilla have been noticed by other people also. http://www.sti.uniurb.it/padovani/PhDThesis/main.pdf Extract from page 195, {BLOCKQUOTE Performances of MathML rendering are not acceptable in some cases. Because there is just a minority of documents with embedded MathML markup, it is hard to say how representative these cases are, and what will be the typical mathematical document on the Web in the future. For sure, Mozilla is not an option for the mathematical library provided by HELM and MoWGLI. These days, on a high-end machine, an average theorem in the HELM library takes from 10 to 20 minutes to render in Mozilla. The problem is not due to pathological situations like trashing: an investigation made together with the author of MathML support in Mozilla revealed that this is most likely due to a) the nature of the MathML documents in HELM, which have lots of nested tables, and b) the implementation of MathML tables by means of the HTML table layout algorithm. The same documents rendered by GtkMathView take from 2 to 3 seconds, also achieving a higher quality rendering (using the same fonts). } If everything is working fine for you David, then that is a good thing. Juan R. González-Álvarez Center for CANONICAL |SCIENCE)
Received on Monday, 14 April 2008 13:01:35 UTC