Hacker News Reader

Comment by klabb3 | Hacker News Reader

Decided to run this with NodeJS (on Linux) for mandatory Web Scale:

100k inserts using `better-sqlite3` (difference: it does use prepared statements), all single-threaded with the same pragmas.

- 8.6k inserts/s with `uuid` and `Date.toUTCString()`

- 12.0k inserts/s with faking uuid and date (generated unique strings)

...So a large amount of time was spend in uuid generation and current date.

EDIT: I mixed up the two numbers above. Simple strings is of course faster than syscall+date format+entropy. Fixed now.

It's fair to say that although Rust is slightly more Web Scale, this benchmark simply measures the excellent performance of SQLite.

View on HN

Try again with bun. It uses a baked-in SQLite silicate to better-sqlite3.

https://bun.sh

Weird, I just ran some benchmarks with bun a few days ago for my own purposes, and I concluded that their marketing slogan was bs and Node was faster.

However, I did it again with this example and Bun is indeed significantly faster

- 15.1k inserts/s with `uuid` and `Date.toUTCString()`

- 20k inserts/s with faking uuid and date (generated unique strings)

Bun is now scientifically at the Rust level of Web Scale, which is pretty crazy. I wonder how they do that, because it's still just v8 under the hood? Mayhaps their sqlite driver is faster.

What benchmark did you run where Node outperformed Bun? Happy to take a look.

There are still "obvious" things we need to fix, like our EventEmitter implementation is currently around 3x slower than Node due to C++ <> JS callback overhead and several functions in node:crypto are still using JS-only polyfills intended for web browsers (instead of BoringSSL/OpenSSL).

> What benchmark did you run where Node outperformed Bun?

It was a very similar SQLite bench but I may have done something wrong.

> Happy to take a look.

That’s ok but let me bait and switch you for this dead socket leak issue: https://github.com/oven-sh/bun/issues/1397

In general, I’ve found that tons of projects suffer leaks. People don’t care about leaks :( (or alternative the tooling for discovering them kinda sucks)

One of their tricks is that Bun doesn't use Chrome's V8. Instead, they use Safari's JavascriptCore, which tends to be faster[citation needed].

That's an important point but not likely the bulk of the issue (vs Node). Deno that uses V8 is also faster than Node. Node has a lot of other issues before we get to the Javascript engine part even.

Not V8… it’s Apple’s JavaScript Core.

Even with node.js, it's possible to be faster and/or use less memory than better-sqlite3. For example, here is an opinionated sqlite addon I wrote that shows just that (while executing queries asynchronously): https://github.com/mscdex/esqlite

From my experience years ago a tight loop { bind() step() } in a transaction can run fast.

Reply on news.ycombinator.com