keynha

6 ms·

1.
by keynha 2mo ago
Namespaces, cgroups, mounts, and seccomp account for only 3ms of your 30ms cold start, and exec plus runtime init is 22ms. Is that 22 about the floor for full Node compatibility, or is there still room to cut it?
2.
by keynha 2mo ago
How do you decide when the cache-hit savings are worth riding out a degrading route versus eating the cold start on a healthy one?
3.
by keynha 2mo ago
Makes sense, thanks. The fixed 10-per-bucket cap is what keeps every viewport the same size no matter the zoom, that's the elegant bit.
4.
by keynha 2mo ago
This is really nicely done, the log zoom makes 4M events feel navigable instead of like a wall. How are you serving a viewport at that scale: precomputed levels of detail, or a query against the raw events on each pan and zoom?
5.
by keynha 2mo ago
0.84 Spearman fidelity to the MiniLM teacher at ternary precision is a striking result. How much of that is the quantization-aware training doing the work, versus what a post-training ternary quant of the same encoder would give you?
6.
by keynha 2mo ago
Dropping the explicit P_DIRTY flag in 1.0 is a neat change. What tracks which pages still need to be written back at commit now that the flag is gone?
7.
by keynha 3mo ago
The number that jumped out is 9GB restoring in 2.25s from S3 but 9s from local NVMe. I'd have bet on local, so the inversion is surprising.
8.
by keynha 3mo ago
That coexistence is also why the GC-free rewrite helps more than the speed numbers suggest. An archiver is allocation-light until it hits a compression burst, then the Go heap can spike toward 2x live right when Postgres wants that memory.
9.
Riskratchet: Stop AI-generated code from rotting your codebase(github.com)
2 points by keynha 3mo ago|0 comments
10.
by keynha 3mo ago
Fiu was told not to reply and had no tools wired up, so the only way it could lose was by printing the secret straight back, which is the half models are already trained hard to resist. The case worth testing is when the agent can send mail
11.
by keynha 3mo ago
Headless, so there is no screen to composite and no GPU passed into the VM. Firecracker has no GPU passthrough, so GL work falls back to SwiftShader, the software rasterizer. For automation that is fine. The cost is in layout, JS and networ
12.
by keynha 3mo ago
For most apps the backend is slower than the proxy by a wide margin, so Caddy is nowhere near the bottleneck. Where it flips is high connection churn, since TLS handshakes are the expensive part and a flood of short lived connections withou
13.
by keynha 3mo ago
The reason to use it is that it skips the double lookup. A normal rowid table with a UUID primary key keeps two B-trees: the table itself keyed by the hidden rowid, and a separate index from your UUID to that rowid. A lookup by UUID walks t
14.
by keynha 4mo ago
In LLM serving, I treat the failure mode at the end of this (long-lived blocks interleaved with short-lived ones, which expandable segments still can't merge across) as the steady state, not an edge case: weights and graph buffers sit
15.
by keynha 4mo ago
Behavioral signals are the usual answer: risk-scored, invisible challenges; proof-of-work (cost without identity, though it taxes mobile); and signup-velocity/rate limits that stop cheap abuse before any challenge fires. The reason fin
16.
Optimizing Our ML Feature Store: Cutting Compute Costs(kayhan.dev)
2 points by keynha 4mo ago|0 comments
17.
A maintainability ratchet for AI-assisted Python(kayhan.dev)
1 points by keynha 4mo ago|0 comments
18.
Five LLM agents play Werewolf in-browser, each with a private DuckDB(kayhan.dev)
2 points by keynha 4mo ago|0 comments
19.
Arrow Flight vs. JSON in Next.js: Benchmarking Python and Go(kayhan.dev)
1 points by keynha 4mo ago|0 comments