Highlights from Twitter Spaces discussions 2021.

Subscribe to my newsletter, support me on Patreon or by PayPal donation.

In this short blog post, I decided to capture the most important highlights (for me) from all the Twitter Spaces conversations that I had during the year 2021. Some of those are not exact quotes, but rather my interpretation of their thoughts (I hope they call me out if I skrewed it up). Everything in this post is in chronological order as episodes were coming out. Recordings of all the episodes are available on my youtube channel (sorry about the mediocre audio quality).

Daniel Lemire:

• “Benchmarking should become as common as unit tests.”
• Think about SW performance from the first principles. Ask yourself: “how fast my code COULD have run”.

Nadav Rotem:

• “Right now LLVM compiler optimizes C++ as if it was C (pointers all over the place).” We could benefit from capturing high-level language semantics as well.

Mark Dawson:

• Low-latency industries (high-frequence trading, real-time robotics) fight hard the fact that most HW and SW components are built around maximizing throughput, not latency.

Ivica Bogosavljevic:

• “Memory bottlenecks are the most common performance issue in modern SW.”

Thomas Dullien:

• “I predict that performance optimization and efficiency in the cloud is going to be way more important 5-10 years down the line to the point where we will eventually have the equivalent of the C-level executive in organizations that’s responsible for the efficient delivery of digital goods.”

Tomer Morad:

• “The world doesn’t like to do SW performance tuning”. Systems should be able to tune themselves automatically.

James Reinders:

• “Parallelism is in data.”
• A better algorithm is better than a better HW.
• “SW libraries should be heterogeneous from the start”.
• “We all will be heterogeneous devs in 10 years from now”.

Arnaldo Carvalho De Melo:

• “eBPF is eating the world”.

Bryan Cantrill:

• “People serious about SW do their own HW.”
• The future is in HW/SW co-designed open platforms.

Andrey Akinshin:

• There is no universal benchmarking method/approach that can be applied everywhere. Benchmarking very much depends on your business goals.

Some of the projects that we were talking about:

• simdjson - one of the fastest JSON libraries capable of parsing gigabytes of JSON per second.
• Prodfiler - whole-system profiler that allows you to continuously monitor performance of your cloud applications.
• Concertio - static and dynamic AI-powered infrastructure optimizer.
• DPCPP - standard C++ and SYCL LLVM-based cross-architecture compiler.
• OxideComputer - a startup that focuses on building hardware, with the software baked in, for running infrastructure at scale.

If you were my guest and you don’t see yourself on this list, sorry, I haven’t saved the notes for every episode, so it’s hard for me to recover everything. Sorry about that.