|
# Dissecting MathJax Performance
|
|
> This is working draft.
|
|
|
|
|
|
This posting gives an overview over the different aspects that affect MathJax performance.
|
|
This posting gives an overview over the different aspects that affect MathJax performance.
|
|
|
|
|
... | @@ -164,7 +164,11 @@ A very basic problem is that the Knuth-Plass bottom-up algorithm we use has to w |
... | @@ -164,7 +164,11 @@ A very basic problem is that the Knuth-Plass bottom-up algorithm we use has to w |
|
|
|
|
|
We could investigate a fundamentally new approach, letting the browser do the layout for us. The latest CSS modules such as flexbox could enable native rendering speed while offering much improved rendering.
|
|
We could investigate a fundamentally new approach, letting the browser do the layout for us. The latest CSS modules such as flexbox could enable native rendering speed while offering much improved rendering.
|
|
|
|
|
|
|
|
#### Dropping support for legacy browsers
|
|
|
|
|
|
|
|
An big questions is how much support for legacy browsers, in particular IE<9, is holding speed back. Browser JavaScript engines changed the way of optimizing javascript execution.
|
|
|
|
|
|
|
|
Caveat emptor: this would probably lead to a re-write of much of MathJax.
|
|
|
|
|
|
## Remarks on speed
|
|
## Remarks on speed
|
|
|
|
|
... | @@ -195,6 +199,7 @@ Chrome 29: ~1.7sec |
... | @@ -195,6 +199,7 @@ Chrome 29: ~1.7sec |
|
Windows 8 / IE 10: ~2.5sec (virtualbox + Microsoft's free testing VM)
|
|
Windows 8 / IE 10: ~2.5sec (virtualbox + Microsoft's free testing VM)
|
|
Windows 7 / IE9: ~3sec (virtualbox + Microsoft's free testing VM)
|
|
Windows 7 / IE9: ~3sec (virtualbox + Microsoft's free testing VM)
|
|
Windows 7 / Firefox 22: ~3.8sec (virtualbox + Microsoft's free testing VM)
|
|
Windows 7 / Firefox 22: ~3.8sec (virtualbox + Microsoft's free testing VM)
|
|
|
|
Windows 8.1 / IE11: ~3sec (virtualbox + release candidate)
|
|
|
|
|
|
|
|
|
|
A copy of the page with TeX converted to MathML was ~0.5-1 sec slower. |
|
A copy of the page with TeX converted to MathML was ~0.5-1 sec slower. |
|
|
|
\ No newline at end of file |