Re: Safari 4 and Chrome super slow with charts, Firefox is fine

[Eugene Lazutkin]

Thank you for the interesting information. I'll look into all this
"string concatenation" and "toString" business hopefully optimizing some
things. Meanwhile, can you provide a super-simple sample that
demonstrate the problem? Probably the best idea is to open an
enhancement ticket in the trac and attach a sample to it.

Additionally I suggest to open a ticket with webkit attaching the same
sample. Usually it helps if you can host it somewhere, so they can just
click on the link and see the problem for themselves. I can monitor this
ticket as well if you include its link in our ticket.

The reason for opening a ticket with the webkit team is that to date
nobody complained about its JavaScript performance, or GC performance,
or string performance --- most probably it is compounded by their
implementation of SVG, like suboptimal internal data structures that
don't scale up well. We already had some performance problems traced to
their use of linear lists rather than spacial trees (logarithm vs.
linear complexity). It may be something like that.

Thanks,

Eugene Lazutkin
Dojo Toolkit, Committer
http://lazutkin.com/

On 08/11/2009 01:38 PM, Nick Mitchell wrote:

> Hi everyone!
>
> We are using stacked bar charts, and have found that with around 250
> data points and 5 series, Safari 4 (and Chrome!) totally bog down.
> Firefox 3 and 3.5 render our page in about 3-4 seconds, whereas Safari
> takes upwards of 20 seconds, including some major time spent
> "beachballing". Chrome, a build of 3, takes much longer than that.
>
> When i take an MacOS Activity Monitor sample, I see that it is caught
> in some combination of GC, mmap, munap, FastMalloc and FastFree, and
> etc. These all seem to be caused by string concatenation coming from
> inside the dojo rendering code. Chrome profiles show similar activity,
> though the build of Chrome seems to have far less debugging
> information, so it's harder to tell.
>
> Has anyone experienced this? Are we "coding crazy", or is this a
> WebKit issue? Firebug profiles show plenty of calls to toString(),
> almost all of which are internal dojo things. This preponderance of
> strings does corroborate the information we have gleaned from the
> Activity Monitor profiles.
>
> This is with a build of Dojo 1.3.1 -- we made our own optimized build,
> to (dramatically) help page load times.
>
> thanks!
> nick mitchell
> _______________________________________________
> FAQ: http://dojotoolkit.org/support/faq
> Book: http://dojotoolkit.org/docs/book
> Forums: http://dojotoolkit.org/forum
> Dojo-interest@...
> http://mail.dojotoolkit.org/mailman/listinfo/dojo-interest
>

_______________________________________________
FAQ: http://dojotoolkit.org/support/faq
Book: http://dojotoolkit.org/docs/book
Forums: http://dojotoolkit.org/forum
Dojo-interest@...
http://mail.dojotoolkit.org/mailman/listinfo/dojo-interest