This article outlines how to create a realtime heatmap of your syscall latency using HTML5, some great node modules, and DTrace. It was inspired by talk that Bryan Cantrill and Brendan Greg gave on Joyent's cool cloud analytics tools. While specific, the code provided could easily be adapted to provide a heatmap of any type of aggregation Dtrace is capable of providing.
System Requirements
First thing's first, you're going to need a system with DTrace. This likely means Solaris (or one of its decedents), OS X, or a BSD variant. There doesn't appear to be Dtrace available for Linux.
Security
Secondly, please be aware that at the time of writing the demo code contains a fairly substantial secruity vulnerabilty. Namely the d script is sent from the client with no authentication whatsoever. If you bind to localhost this shouldn't be a big deal for a demo. Time permitting I intend to clean up the code.
Dependencies
For this tutorial you'll also need:
node - http://nodejs.org/#download (duh)
npm - https://github.com/isaacs/npm (makes installing modules a breeze)
node-libdtrace - https://github.com/bcantrill/node-libdtrace (provides dtrace functionality)
Socket.IO - 'npm install socket.io' (web sockets made easy)Server
Now we're ready to start writing our web server:
Client
In order to display our heatmap, we're going to need some basic HTML with a canvas element:
Finally the JavaScript client which translates the raw streaming data into pretty picture:
Run It!
Run Heat Tacer with the following. Note, sudo is required by dtrace as it does kernal magic.
sudo node heat_tracer.jsIf all goes well you should see something a moving version of something like the image below.
Contribute
You can find the latest version of Heat Tracer here. It is my hope that this article will provide the ground work for a much more abitious performance analytics project. If you're interested in contributing please let me know.
Further Research
More information about Bryan and Brendan's demo can be found here.
Socket.IO can be found here.