weblwabl @weblwabl

present day present time Joined March 2026

Tweets

21
Followers

15
Following

395
Likes

681

weblwabl @weblwabl

4 weeks ago

@BenjDicken I’ve always thought systems performance was more of a reference than a book to read cover to cover? Is that the case? Or am I horribly mistaken and I should read it 😭

0 0 0 92 0

View Details

weblwabl @weblwabl

2 months ago

Moments where I can do this makes my mind feel so free.

ThePrimeagen @ThePrimeagen

2 months ago

I am remembering right now for my own sanity. There is so much peace in deep thought instead of some insane frenetic agent orchestration fever dream

180 153 5K 341K 320

0 0 3 37 0

View Details

weblwabl @weblwabl

2 months ago

@beffjezos Nightmare blunt rotation

0 0 2 34 0

View Details

@massimo_san @Adriksh It’s not just files though, it’s anything that has a file descriptor associated with it which can include sockets too. Generally databases especially high frequency analytics databases can easily run in to this.

0 0 3 150 0

View Details

weblwabl @weblwabl

2 months ago

@TrisH0x2A Also if you’re doing direct io you should open files with the O_DIRECT and O_SYNC flag. Since you’re not fsyncing something like this is not guaranteed to write to disk.

0 0 2 31 0

View Details

weblwabl @weblwabl

2 months ago

@TrisH0x2A Generally syscalls are expensive due to the user to kernel space context switch. This is why something like io_uring tends to perform well, you batch syscalls before switching from user to kernel space. Not only that it has zero copy buffers.

1 0 2 337 0

View Details

weblwabl @weblwabl

3 months ago

@0xSero New edition of Designing Data Intensive Applications and Database Internals.

0 0 0 60 0

View Details

weblwabl @weblwabl

3 months ago

@buggirl Marc Andreessen if you smooth down his head a bit

0 0 1 447 0

View Details

weblwabl @weblwabl

3 months ago

@_TylerHillery I recently started working through Math Academy as well for the past two weeks. It’s been really helpful. I love the leveling system 😆

0 0 2 715 0

View Details

weblwabl @weblwabl

4 months ago

@sudo_goreng If you want to deep dive in to LSM compaction (definitely relevant to RocksDB). This is an excellent survey paper on the topic: vldb.org/pvldb/vol14/p2…

0 0 1 20 1

View Details

weblwabl @weblwabl

4 months ago

@johncrickett Database Internals and Designing Data Intensive Applications are two of my favorites I’ve read in the past. Right now I’m reading through Understanding Deep Learning.

0 0 1 18 0

View Details

weblwabl @weblwabl

4 months ago

@GeoffreyHuntley I’ve been thinking about checking out ocaml for some time, I noticed that Jane Street has a “learn ocaml” repository, but it’s several years old: github.com/janestreet/lea… I was thinking about going through it but I wonder if there are better options.

0 0 0 7 0

View Details

weblwabl @weblwabl

4 months ago

@_trish_xD Reading OSTEP is generally a good starting point.

0 0 0 11 0

View Details

weblwabl @weblwabl

4 months ago

Everything this dude posts is an absolute gold mine.

Vivek Galatage @vivekgalatage

4 months ago

Another periodical repost of this amazing course from MIT.

1 21 287 10K 308

0 0 9 83 0

View Details

weblwabl @weblwabl

4 months ago

@blind_via @suyogvyawahare I’d join a software people interested in hardware discord.

0 0 0 10 0

View Details

weblwabl @weblwabl

4 months ago

@josephdviviano Is Claude okay 😭

0 0 1 36 0

View Details

weblwabl @weblwabl

4 months ago

@trashh_dev “Take a step back, think through some things, and then come back to your keyboard. Ping me when you’re ready for re-review.”

0 0 0 3 0

View Details

weblwabl @weblwabl

4 months ago

@claudeai Why wouldn’t I just have claude code run ‘gh pr diff <pull request #>’ locally? This seems like a dud idea when I already have the tooling to do it.

0 0 0 8 0

View Details

weblwabl @weblwabl

4 months ago

@lcamtuf I tend to use github.com/DrCatHicks/lea… when I want follow up questions from Claude. Most recently used it while learning about gradient descent. I also will often go out of my way to ensure Claude does not explicitly solve problems for me but instead act like a fancy rubber ducky.

0 0 1 3 0

View Details

weblwabl @weblwabl

4 months ago

@KatanaLarp Many people are highlighting that as a follow up you can use the models to improve performance. This is true. But also, I think that once a code base gets to a specific complexity it is very tedious to drive performance gains via targeted prompts and profiling. Great read!