Specbranch
specbranch.com · Personal blog · on HN since 2022
Specbranch is a blog on low-level performance, hardware, and engineering failures, with posts on servers, floating-point random numbers, lookup tables, and Optane.
What HN saysHN commenters link it as a source on topics like single big servers, lookup table tradeoffs, and random floats; one calls it a good set of posts.
23HN people shared it
9linked it in comments
29stories submitted
2.3Kstory points
3 months agolast shared
2022people per year2026
Most-shared posts
- 1Use one big server5 people · 1.8K points · first shared 2022 · 585 comments on HN
- 2You (Probably) Shouldn't Use a Lookup Table4 people · 9 points · first shared 2022 · 4 comments on HN
- 3The Knight Capital Disaster: losing 440M in 45 minutes (2023)4 people · 8 points · first shared 2023 · 1 comments on HN
- 4Rest in Peace, Optane3 people · 10 points · first shared 2022 · 2 comments on HN
- 5Perfect Random Floating-Point Numbers2 people · 115 points · first shared 2025 · 61 comments on HN
- 6Abstraction Is Expensive2 people · 7 points · first shared 2022 · 1 comments on HN
- 7The Most Useful Statistical Test You've Never Heard Of2 people · 4 points · first shared 2022 · 1 comments on HN
- 8Division of 8bit nums in C is faster than a DIV instruction on Intel Haswell CPU2 people · 4 points · first shared 2022 · discussion on HN
- 9Five Nine Problems2 people · 3 points · first shared 2024 · discussion on HN
- 10Time Programming for Lawyers and Jurors2 people · 2 points · first shared 2024 · discussion on HN
- 11Python Is Like Assembly2 people · 2 points · first shared 2022 · 1 comments on HN
- 12Teach your kids bridge, not poker1 people · 327 points · first shared 2022 · 223 comments on HN
- 13What Happened with FPGA Acceleration?1 people · 4 points · first shared 2022 · discussion on HN
- 14Less-Than-Linear Fibonacci1 people · 3 points · first shared 2022 · discussion on HN
- 15Performance Numbers Worth Knowing1 people · 3 points · first shared 2022 · discussion on HN
What HN people wrote when they linked it
pclmulqdq · 2023-07-06
I kind of fell off the wagon after a strong 2022, but there are a few good ones:
https://specbranch.com
datadrivenangel · 2025-11-12
The powerpeg feature flag had been deprecated for 9 years? In aggregate, yes, the engineering led to the disaster, but the specific engineers had been making everyone piles of money for a long time. The mistakes were fatal, but in a tiny amount of time on a system that old it's honestly surprising it didn't happen sooner.
https://specbranch.com/posts/knight-capital/
Nemo_bis · 2025-10-29
Reminds me of https://www.specbranch.com/posts/one-big-server/
HeliumHydride · 2025-06-02
This article shows how to generate perfectly-distributed random floats: https://specbranch.com/posts/fp-rand/
andy99 · 2024-05-15
Why you (probably) shouldn't use a lookup table https://specbranch.com/posts/lookup-tables/ gives some discussion about when it's appropriate generally. My narrow experience is that you can do a lot of real time calculation before it's faster to do a lookup.
akvadrako · 2024-03-03
If that is faster would seem to depend on if you can get most lookups from the L2 cache. Otherwise you're waiting for main memory, which is a few hundred cycles. Even with multiple loads in parallel, it would be hard to beat arithmetic.
https://specbranch.com/posts/lookup-tables/
cosmic_quanta · 2023-12-04
Indeed, I'd rather have the program crash rather than repeat this nightmare: https://specbranch.com/posts/knight-capital/
renegat0x0 · 2023-07-12
I am running a set of SBCs, but I read that it is better to use one server [1]. I suppose one could go with a VM, or portainer, or Yacht, or some other software to visualize everything.
Right now I am using a set or solutions:
* email * [2]
* files * using NextCloud
* blog * using Hugo
* news * using my own RSS reader
* pihole * to filter things
What I consider to use in the future
* pictures * using photoprism
* citadel * I found some sort of suites like citadel. Not sure if that works, if it is good, or if it fits my needs at all [3]. At the first glance seems like a fun project.
Links:
[1] https://specbranch.com/posts/one-big-server/
[2] https://github.com/LukeSmithxyz/emailwiz
[3] https://www.citadel.org/
All in all it is time consuming to setup up, and update regularly.
kragen · 2022-08-02
It's not unlimited scale, it was just over 700_000 people according to https://news.ycombinator.com/item?id=32320697. If your web page is 100 kilobytes, which seems like the maximum that would be reasonable, that's 70 gigabytes of traffic. AWS charges US$0.09 per GB so this is US$7. If you eliminate the cloud premium it's closer to US$1.
In fact loading their home page takes 3.1 MB for me over two minutes. Searching for a flight (AAL301 in this case since SPAR19 has already landed) brings this to 3.8 MB. At this size it would be 266 gigabytes, US$27. Reloading the page, I see that this was about 180 HTTP hits, though a lot of those were ads (I guess they have a way to monetize the traffic) which were blocked by uBlock Origin.
But doesn't it take a lot of CPU and RAM to serve 700_000 page views, especially at 180 hits per page view? That's 126 million hits, after all! Well, of course you can write your code arbitrarily inefficiently. According to https://crozdesk.com/software/fastly/pricing Fastly charges US$0.0075 per ten thousand hits, so if you could serve all those hits from Fastly it would cost you just under US$100. And probably if you're getting 700_000 people looking at the same thing you should figure out how to make all those hits cacheable either in Fastly or in something slower but cheaper. This probably isn't the first popular flight on FlightRadar24, even if it's the first one that's this popular.
(Also though you probably don't need 180 hits to…