Makes things fast. Expert in computer performance and eBPF. Intel Fellow, ex-Netflix. Created flame graphs. Aussie. Opinions my own. brendangregg
@aus
.social
My BPF book is publishing today as an eBook (it's already on Amazon), and InformIT have a 55% off for 2 or more Black Friday sale until Dec 3rd! Get copies for your whole team (thank you). Code "BF2019".
#BPF
Good news from Oracle: DTrace is GPL'd (thank you!). I'd guess by the end of 2018 we'll have a working /usr/sbin/dtrace on Linux for running D scripts (using libbcc+eBPF on the backend)
Thanks, Greg! I'm thrilled to be joining at this exciting time. I'm looking forward to improving the performance and observability of everything -- all xPUs, apps to metal -- working with great people at a company I have long respected and admired.
Welcome
@brendangregg
to our software team
@Intel
. Having worked with him at Sun Microsystems, his contributions to improving systems performance needs little introduction. He will focus on strengthening our leadership in computing across Intel’s xPUs.
New startups are developing eBPF as a "zero instrumentation" observability/APM solution. Trace plaintext HTTP/gRPC/etc. across your site calls without code changes (zero added instrumentation). Currently via uprobe hacking. 1/n
I'm hiring for my first ever team. My first job description is for Cloud Performance and eBPF engineers! (many locations, I don't know why it defaults to saying Arizona):
Did Linux get slower after 4.14? Yes, if you use the KPTI defaults. How much? Between 1 and 800%, depending on your workload. I expected our workloads to slow by between 0.1 and 6%. I wrote about it before
"a regular expression that backtracked enormously and exhausted CPU" -- don't use regexps in perf critical code (as
@guycirino
would insist); thanks
@Cloudflare
for the writeup
eBPF in Microsoft Windows is a big deal, not just for Microsoft, but for us in BPF. Technical decisions with BPF will consider other kernels when appropriate, so things work everywhere and not just on Linux... 1/2
Welcome kubectl-trace by
@fntlnz
to
@iovisor
, the Linux Foundation eBPF project that hosts bcc and bpftrace. kubectl-trace runs bpftrace on Kubernetes.
New post: Computing Performance: On the Horizon. My
@USENIX
#LISA21
plenary session slides and video are online. Summarizes the present and my predictions for the future of performance.
The BPF Performance Tools book immediately sold out on Amazon (thanks for getting it!) I think there are still copies on InformIT while they print more
eBPF comes to applications: BPF Memory Cache (BMC) is an eBPF memcached accelerator. Currently for GET requests over UDP. Promising work! Thanks
@Orange
Don't assume any eBPF observability post/talk/book has had any review by the eBPF community, or is based on any up to date research. I saw another case of 5-year old advice posted as new. Most advice (2021) summarized in this one slide