[rabbitmq-discuss] Experimental RabbitMQ Topology Visualiser

[Matthew Sackman]

On Tue, Jul 05, 2011 at 05:42:08AM -0700, Joseph Marlin wrote:
> Speaking of CPU usage, I'm getting a *lot* of it when I have the tab with 
> this visualizer visible in Chrome on Ubuntu: http://i.imgur.com/uhU2m.png

Yup. n-body simulations are CPU intensive, especially when you're
crippled by a language like javascript. That said, firstly, since the
release at the end of last week, I've done further optimisations such as
frustum culling, and secondly, profiling in chrome (it's built-in
profiler is very very good!) shows that actually the CPU cost is not in
the physics, but in the canvas drawing operations - the general state of
canvas implementations seems to be fairly unoptimised, especially when
it comes to things like drawing text. This could be because most of the
shiny shiny demos don't spent a lot of time drawing text or doing much
other than moving balls attractively around a screen ;)

> Is that normal? The dip in the beginning was when I switched to a different 
> tab for a few seconds.

Alas so. I'm doing what I can to minimise this, but it's been fairly
well demonstrated that implementing something in flash still leads to
better performance than an HTML canvas on modern browsers - they still
have some way to go yet to close the performance gap.

Matthew