3 ms·
Linux caught up last year: http://www.brendangregg.com/blog/2016-10-27/dtrace-for-linux-2016.html http://www.brendangregg.com/blog/2016-10-27/dtrace-for-linux..
by brendangregg 9y ago
Linux caught up last year: http://www.brendangregg.com/blog/2016-10-27/dtrace-for-linux-2016.html http://www.brendangregg.com/blog/2016-10-27/dtrace-for-linux...
Thanks for getting the book! DTrace is invaluable for solving a set of needs, and we're using it on our FreeBSD CDN. Over time, Linux has been adding capabilities that can solve the same set of needs, although using different technologies.
Right now, the Linux kernel (4.9+) can do the same things as DTrace, but the interface we're using (bcc/BPF) is currently 10+ times more verbose and much more difficult to use. Who does that hurt the most? At Netflix, me, as it's my job to add advanced observability to our analysis GUIs, and actually write the bcc/BPF.
So far I've given conference talks on bcc/BPF at Performance@Scale, USENIX LISA, linux.conf.au, BSidesSF, and SCALE. Next is Velocity Conf.
- srcmap 9y agoBrendan Love your work on DTrace. Have you seen / configured a system that keep the FTRACE/DTRACE log data in memory during a crash/ reboot cycle for postmortem kernel crash type analysis? If so, love the see any reference to how to setup something like that for typical x86 Linux. Did that long time (10+years) ago on MIPS+vxwork/Linux with someother trace log type system. Miss that capability..... Thanks