Why Linux people lust after DTrace (and how DTrace saved Twitter)

You haven't suffered until you've experienced a Sun sales team desperately rant for half an hour about how DTrace makes Sun a relevant technology company again.

While I'm sure that DTrace does nifty things, this was probably the flimsiest explanation of what makes it great that I've read. The summary seems to go like this:

There's this great tool for Solaris called DTrace. It's only available on Solaris, and it's so cool. One time, Twitter found a problem with it that increased speed by 30%!!!

DTrace is very useful for certain tasks that are traditionally hard to do on Unix. Mainframe OSes for example make it easy to figure out who is doing the most disk IO etc. while under Unix this is tough. DTrace allows for enough visibility to figure out most things that a Mainframe OS would let you find out.

I've watched a few lectures from the authors of DTrace and I've been nothing but impressed. Especially with the Python ad Ruby instrumentation. Being able to instrument ANY OS call with zero off overhead is pretty useful.

It's a shame that (for some reason) it's crippled on OS X and the FreeBSD port is progressing so slowly (my dev and deployment platforms).

My understanding is that Joyent, Twitter's host at the time, helped them significantly with the dtrace stuff.

i wouldn't go so far as to say it saved twitter

with some elbow grease, systemtap + kprobes on linux is not a bad combo at all.

edit: comparison of systemtap and dtrace feature sets http://sourceware.org/systemta/wiki/SystemtapDtraceCompariso...