HackerTube

"Performance (Really) Matters" with Emery Berger

Association for Computing Machinery (ACM)3 HN recommenders · 2022–2026
7K views · talk · evergreen · Uploaded 2020-10-28

Emery Berger argues that performance is hard to measure and deliver on modern systems, showing how code layout and link order can swamp optimization effects. He presents tools that evaluate performance more reliably.

HN points to it whenever benchmarking pitfalls and layout effects come up, as an eye-opening talk on how unreliable naive measurements are.

Software performance & optimizationCompilers, interpreters & program analysis

Watch on YouTube

7 comments from Hacker News

@kaliszad2022 · in ““Fast Kernel Headers” Tree -v1: Eliminate the Linux Kernel's “Dependency Hell””

I know, and that is exactly my point. It isn't like "send to REPL" in pretty much any half-decent LISP. And it hurts development comfort, speed and maintenance is way more burdensome than what it could be. Here, you would need to involve a compiler of course and do more due diligence and then run ftrace or something to exchange the code. Also, you would probably in some cases need to make sure, the ordering of code is right as to not hurt performance. (See Emery Berger's talk [VIDEO] which probably applies here as well.) Of course, with great power comes great responsibility. You could add some optional checking facility, that would be able to roll back the changes to code if some conditions were met.

CitesView on HN
@sitkack2022 · in “The true cost of linked lists”

It isn't CS 101. You are being overly dismissive and even CS people from top schools get this stuff wrong. The courses that introduce big-O are often the weeder courses, they don't aim to education, they aim to flunk out the people that might be hard to teach so that CS departments can gate keep. When you have top CS researchers having to give talks to remind people that layout matters more than instruction selection, the CS as a whole has focused on the wrong things. [VIDEO] Figure how to build people up, not tear them down.

CitesView on HN
@vanderZwan2022 · in “Fast memcpy, A System Design”

> The main trap is trying to benchmark and profile it without the context of the whole application. Emery Berger's "Performance (Really) Matters" talk also goes into this[0]. He paints a really, really depressing picture of how unreliable benchmarks truly are. Like, things like link order having ridiculous impacts and stuff like that. [0] [VIDEO]

RecommendsView on HN
@nerpderp822023 · in “Four Kinds of Optimisation”

Performance (Really) Matters by Emery Berger talks a lot about layout. [VIDEO] And the topic of Coordinated Omission, "How NOT to Measure Latency" by Gil Tene [LINK] Injection slowness into an application to measure the relative impact of that portion of the application has really worked well for me.

CitesView on HN
@reitzensteinm2024 · in “When Optimising Code, Measure”

If you're measuring performance and haven't seen this talk by Emery Berger, I highly recommend it. It's mind blowing, but also terrifying when you realize how many mistakes you've been making up until this point. [VIDEO]

Strongly recommendsView on HN
@sitkack2025 · in “Disabling Zen 5's Op Cache and Exploring Its Clustered Decoder”

If you enjoy the science of injecting slowness to determine which component has the largest impact on performance, you would enjoy this work by Emery Berger. Coz: Finding Code that Counts with Causal Profiling [LINK] "Performance (Really) Matters" with Emery Berger [VIDEO]

RecommendsView on HN
@vanderZwan2026 · in “To study how chips work, MIT researchers built their own operating system”

I think you're better off watching any of the various recordings "Performance Matters" conference talks[0] by Emery Berger to learn about all the ways that developers are benchmarking the wrong thing using the wrong techniques, and what you could do instead [0] [VIDEO]

RecommendsView on HN