Top all-time · #5530
Marton Trencseni
Personal blog · bytepawn.com · on HN as Maroprogrammingscience
Marton Trencseni writes Bytepawn, a personal blog at bytepawn.com about programming and science. Since 2008, 14 different Hacker News readers have linked it 85 times, most often “Readings in Distributed Systems”. It is still writing: the last post was on 2026-06-11, and you can follow it by feed.
Too few sampled comments (4) to say how often HN recommends it.?
Why HN links it
HN readers link Marton Trencseni's posts on data and distributed systems, including an ETL tools comparison (Luigi vs Airflow), a distributed systems reading list, and a post on Shackleton's job advertisement.
Summarised by Claude from a sample of 4 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- Readings in Distributed Systems4 people · 6 mentions · first linked 2008-10-28
- Luigi vs. Airflow vs. Pinball3 people · 3 mentions · first linked 2016-02-06
What Hacker News says?
http://bytepawn.com/ Recently I wrote about A/B testing topics (lots of Monte Carlo simulations). Before that Deep Learning stuff.
Comment linking /
on “Ask HN: What is your blog and why should I read it?”Submitted Useful Python decorators for data scientists as a story · 233 points
HN discussionI think most of the things people know, it's covered in books like Lean Startup, etc. It's a bit like eating well and working out. We all know we should do it, but most people don't actually eat well and work out. Then, when you see somebody who does it and looks great, you ask them, "What's your secret?". But it's not a secret, it's just that most people don't do it, because it's hard :) One story: When I was at FB…
Comment linking /hack-hhvm-second-system-effect.html
on “Ask HN: CS papers for software architecture and design?”Submitted Breaking Bell's Inequality with Monte Carlo Simulations in Python as a story · 122 points
HN discussionI wish more people would read 'Hack and HHVM', written by Owen Yamauchi, formerly a member of Facebook’s core Hack and HHVM teams. http://shop.oreilly.com/product/0636920037194.do The hidden lesson for me was that rewriting the code in <new-language> is not the only option. Another option is to slowly improve the language/runtime itself until you've essentially switched it out underneath the application, which is wh…
Comment linking /hack-hhvm-second-system-effect.html
on “The eigenvector of “Why we moved from language X to language Y””Submitted Randomness extractors: making fair coins out of biased coins as a story · 105 points
HN discussionGreat comment. What you say matches my experience, but not entirely. My take: The 2% is a bit low as an avg. for ML work; I would say modeling/ML work ranges from 0 to 20% across the "data team". Most of the work is still data wrangling/engineering. I agree that most companies need "full stack data guys". At my current company, there were 2-3 iterations of previous DS teams who had zero impact and left behind nothin…
Comment linking /automating-a-call-center-with-machine-learning.html
on “Machine Learning Books That Helped Me Level Up”Submitted Five ways to reduce variance in A/B testing as a story · 83 points
HN discussionI wrote a clean C++ version of `wc` for fun, and successfully beat the system version by 10-20% iirc. My goal was to make the code much more readable than the GNU/C version, which is a big spaghetti loop, while maintaining the same semantics (ie. return the same counts). Blog posts: https://bytepawn.com/tag/wc.html Code: https://github.com/mtrencseni/wcpp/
Comment linking /tag/wc.html
on “Wc2: Investigates optimizing 'wc', the Unix word count program”Submitted How I format SQL code as a story · 60 points
HN discussionI looked at the wc implementation, there's no way it's compatible with the GNU tools, how it handles wide characters, which is heavily ties to the idiosyncratic C implementation. I know because I once wrote a wc implementation in modern C++ for my own education, and that's where most of my time and code complexity went. Also, the C versions are devilishly fast, with all flags, with optimized code branches. https://b…
Comment linking /tag/wc.html
on “Unix core utilities implemented in Haskell”Submitted Over 70% of all Porsche vehicles ever built are still on the road today as a story · 53 points
HN discussionI've seen 2 non-trivial rewrites of the mainline software at 2 companies. Non-trivial = 10-100M in yearly revenue. In both cases: - it was promised to take 6-9 months, and ended up taking 3-4 years - it never really finished, the old software had to remain in production (along with the new "rewrite") - in some form or the other, while the rewrite was happening, the company lost its customer focus and/or its ability…
Comment linking /hack-hhvm-second-system-effect.html
on “The True Cost of Rewrites”Submitted Optimal coverage for Wordle with Monte Carlo methods – Part III as a story · 50 points
HN discussionMy blog has tons of articles about A/B testing, with math and Python code to illustrate. Good starting point: https://bytepawn.com/five-ways-to-reduce-variance-in-ab-test...
Comment linking /five-ways-to-reduce-variance-in-ab-testing.html
on “Ask HN: Resources about math behind A/B testing”Submitted Luigi vs. Airflow vs. Pinball as a story · 49 points
HN discussionHere is, a slightly outdated, article that compares several ETL workflow tools http://bytepawn.com/luigi-airflow-pinball.html . Why we choose Airflow was because of the following reasons: * Scheduler that knows how to handle retries, skipped tasks, failing tasks * Great UI * Horizontal scaleable * Great community * Extensible; we could make it work in an enterprise context (kerberos, ldap etc) * No XML * Testable an…
Comment linking /luigi-airflow-pinball.html
on “Airflow and the Future of Data Engineering: A Q&A”Submitted Probabalistic Spin Glass (2022) as a story · 38 points
HN discussionI haven't read the OP, but if you're interested in this stuff, my blog is full of MC simulations, both for frequentist and Bayesian experimentation: http://bytepawn.com/tag/ab-testing.html
Comment linking /tag/ab-testing.html
on “Why Bayesian Stats Needs Monte-Carlo Methods”Submitted SSD Reading List as a story · 29 points
HN discussionIn Physics, the log part comes in when you use the Stirling approximation for large N. Ideal gas: https://bytepawn.com/entropy-of-an-ideal-gas-with-coarse-gra... Physical gas: https://bytepawn.com/the-physical-sackur-tetrode-entropy-of-...
Comment linking /the-physical-sackur-tetrode-entropy-of-an-ideal-gas.html
on “An alternative construction of Shannon entropy”Submitted Lamport's Byzantine Generals Algorithm in Python as a story · 28 points
HN discussionShameless plug: https://bytepawn.com/culture-docs-facebook-netflix-and-valve...
Comment linking /culture-docs-facebook-netflix-and-valve.html
on “Move fast or die: Key startup lessons”Submitted Scalable Web Architectures and Application State as a story · 28 points
HN discussionI have a NoSQL startup and I got started by reading "the Google papers": http://bytepawn.com/readings-in-distributed-systems/ Some code: http://github.com/scalien/scaliendb
Comment linking /readings-in-distributed-systems/
on “Leaving Google”Submitted A/B testing and the historic lift paradox as a story · 23 points
HN discussionDisclosure: I work on Astronomer.io, an open source Airflow platform and SaaS, and also contribute to Airflow. IMO the pros / cons are really relative to what you're comparing it to and what your workflow needs are. We've written some guides on "Airflow vs ___" [1] (currently AWS Glue and Oozie). Feel free to email me (in profile) if you'd like to see a comparison to something else. I like these two posts on Airflow…
Comment linking /luigi-airflow-pinball.html
on “Data Engineering Patterns with Apache Airflow [video]”Submitted Readings in Distributed Systems as a story · 23 points
HN discussionWe base our whole DS infrastructure on Airflow (and Superset): http://bytepawn.com/fetchr-airflow.html http://bytepawn.com/fetchr-data-science-infra.html Airflow is somewhere between good enough and pretty cool, it's based on what we had at Facebook (called Dataswarm). IMO in 2-3 years Airflow will be the de-facto ETL standard, like Hadoop used to be for "Big data". If you're rolling your own ETL at this point, you'…
Comment linking /fetchr-data-science-infra.html
on “Data Engineering Patterns with Apache Airflow [video]”Submitted Doing an Ironman as a story · 22 points
HN discussion
