I didn't know much about perf (https://perf.wiki.kernel.org) before watching this talk. I had no idea that perf supported dynamic tracing in the kernel (around minute 40/slide 57). I'm definitely going to play around with it more this weekend.

I always enjoy Brenden's talks. He really knows his stuff.

Awesome presentation. Brendan Gregg shows a scary amount of expertise here. Very motivating to start using more advanced tools.

I consider the Joyent (Ex-Sun guys) the gods of Unix based operating systems, just reading how they ported KVM from Linux -> Illumos displays their deep understanding of what is going on under the hood. Seriously if you want to be an above average Unix system administrator, consider learning and using Dtrace with Solaris/OpenIndiana as it will improve your troubleshooting skills immensely with *nix in general. I just wish it existed in GNU/Linux (yea,yea Oracle is trying this, but still very incomplete atm), and Linux based SystemTap just does not feel as awesome as Dtrace.

Apple should buy Joyent immediately and put them in charge of the entire Service side of Apple (iCloud, etc.).

There isn't any better way for Tim Cook to spend that amount of money, full stop: Apple is getting it's ass kicked on the server side by Google, and they deserve it, because Apple doesn't have the talent and won't get it because, let's face it, who wants to work at Apple on server stuff that has any kind of skills?

Buying Joyent would change that overnight.

I don't think they'll like the idea of dealing with Apple after they left Oracle. They have better things to do.

Has anyone tried the dtrace port for linux? Any comments on it?

https://github.com/dtrace4linux/linux

I stopped watching after having 5 slides read to me verbatim.

Really? You would throw out a helpful presentation for style points?

Haven't listened to the presentation but the diagram doesn't list this project which I learned about from a Misko Hevery post on the googletesting blog. http://freecode.com/projects/fio

Look forward to watching it.

The talk focused on observability tools, not experimental tools like micro-benchmarks. I probably should have made that more clear (I only mention experimental tools on slide 84).

I've actually been using fio a lot recently, especially since it can do non-uniform random distributions.

Any benefit to fio vs. iozone?

Great stuff. Thanks!

    sudo perf top 

is really great!