shachaf retweeted
I wrote a short summary of the proof of the FLP theorem (an impossibility result about consensus).
_Different Kinds of Darkness_ by David Langford: lightspeedmagazine.com/ficti…
As I understand it TigerBeetle is entirely single-threaded and single-sharded, with small atomic transactions. Is serializability very meaningful in this world? It sounds like a claim you'd make about a sharded database or one with complex/long-lived transactions.
Replying to @TigerBeetleDB @jorandirkgreef
"You can have any isolation level you want... as long as it's Strict Serializability!"
shachaf retweeted
Interesting, @shachaf just called my attention to the field with six elements.
It's kind of funny that the RISC-V fence instruction tries to be very flexible, letting you specify an unusually fine-grained bitset of predecessor and successor things to fence, and yet it's not flexible enough to express TSO, so there's an extra bit for that specific behavior.
I encouraged my friend Peter to write up a point about equity vs. cash compensation that we've discussed many times:
I wrote a new blog post on how equity comp is actually worth more than the same nominal dollar amount of cash, because equity packages are basically a call option on shares struck in your labor, and the optionality part of a call has non-negative value. peter.website/equity-beats-c…
shachaf retweeted
I wrote a new blog post on how equity comp is actually worth more than the same nominal dollar amount of cash, because equity packages are basically a call option on shares struck in your labor, and the optionality part of a call has non-negative value. peter.website/equity-beats-c…
shachaf retweeted
Nice to see multi-producer/single-consumer lock-free queue based on my design used for #iouring in the #Linux kernel:
lore.kernel.org/all/20260611…
The original design for non-intrusive/intrusive MPSC queues from 2008:
groups.google.com/g/lock-fre…
cc @axboe
I liked the group we had at my previous job! Maybe I should try to do something like this locally.
If you're trying to learn about databases, distributed systems, etc, it can be really helpful to discuss papers with other people
I've been attending a systems reading group for the last few months, and I've found that discussing a paper with others hammers it into my brain way more than reading it alone
If you're into this kind of stuff, I'd definitely recommend finding (or starting) a small reading group
(also this article is how I discovered the systems reading group at Microsoft)
shachaf retweeted
Alpha-beta pruning in one sentence:
If you want to compute max(8, min(5, ...), ...), you don't need to compute the rest of the arguments to min, because they won't affect the value of the max.
shachaf retweeted
An AC power grid made way more sense in the 20th century, but switching mode DC->DC converters have gotten *so* insanely cheap, small, and efficient, especially with GaN FETs. Would a DC power grid (say, ~20 kV distribution, ~300 volts in your house) make more sense in 2025?
shachaf retweeted
Fast CASPaxos - a leaderless consensus algorithm
reubenbond.github.io/posts/f…
Note to self (because I keep forgetting these numbers and having to figure them out again):
The original Fast Paxos paper suggests two extremes for cardinality-based quorums:
Classic-minimizing: classic = ⌊N/2⌋ + 1; fast = ⌈3N/4⌉
Fast-minimizing: classic = fast = ⌊2N/3⌋ + 1
shachaf retweeted
Wherein I hope my understanding is correct, otherwise I'm going to be roasted by @filpizlo -- corsix.org/content/simplifie…
In my mind Paxos is less a very specific algorithm and more a collection of ideas for how to make a consensus system. (There are some concrete Paxos algorithms, of course, but you'd rarely use them as-is.) Raft is nearly a point in the space spanned by those ideas.















