Apparently that guy has a habit of doing that. If I were the owner of that repo, I would delete the comment and ignore him. Block him if he does it again. It really should not be necessary to explain to people why such behaviour is bad.

Show thread

Swearing at people in github comments on their commits is not a good way to start a conversation about how the code could be improved.

Linux server at home is only connected to wifi and power. No screen or keyboard. So of course it won't come back online after a software update.

Chris Vest boosted

Welp, here we go again.

This client's got events coming in from the future - in this case, just one hour in the future, but still.

Folks, please, please, please set up an NTP server on your local network and configure all your boxen to get their time from -one- place.

Chris Vest boosted

It's taken a few tries to get into writing Rust code. The problem, I think, is that there's a non trivial amount you need to learn about he language before you can make anything of significance. Strings, boxes, ownership, sharing, life time annotations, the module system, etc. You need to know all of these before you start to make progress. And that's too much to learn all at once. I had to start several times, where each time I tried, I started grasping one more concept.

There's a couple of interesting things that would change if you designed a LLAMA like system for NVRAM instead of flash. Question is if an implementation can support both transparently. Would certainly be incredibly useful in Neo4j.

Chris Vest boosted

I guess it's reasonable to assume all Macs have all SSDs, in terms of performance profile assumptions. I need to decide if parallel random IO is a reasonably performant thing to do in a given directory, and as far as I can tell only Linux has the necessary APIs to make a well educated guess.

Chris Vest boosted

The Neo4j import-tool is the fastest data ingestion tool for any OLTP graph database that I know of. I've found a way to make it 50 times faster (in the one particular case that I've tested), and might be able to push it to a 100 fold performance improvement.

Chris Vest boosted

Many comp.sci. graph papers assume that graph elements, vertices and edges, can be counted in 32 bits. It's cute, and certainly convenient when possible, but in industry we need 64 bits (or at least 40 bits - enough to count one trillion) for this stuff.

Chris Vest boosted

@aparrish And this is why in the 24th century, Picard still telling the computer "Tea, Earl Grey, Hot" every time

A funny thing about being so close to the release of Neo4j 3.2 is that all I'm thinking about is he version that comes after. From my perspective, 3.2 is practically done, even though it's not even released, and the buzz around it hasn't even started yet.

The B-tree in @hyc's LMDB is also an interesting thing to study.

Show thread

We had a reading of the original LSM-Tree paper. Our suspicions were confirmed: it's not particularly impressive, and we have no use for it. If you are considering it, you should also take a serious look at fractal trees, dancing trees, generational B+trees and the BwTree with LLAMA, before deciding what to go with.

Really wish MacOS had `perf`. The ecosystem of introspection and observability tools is the only thing I like about Linux.

"The biological pathways from caffeine to code are intricate. Here we see a young male, busily metabolising."

The LLAMA paper - the storage/persistence back-end for the BwTree - was pretty bad at this as well.

Show thread
Show older
Mastodon

Server run by the main developers of the project 🐘 It is not focused on any particular niche interest - everyone is welcome as long as you follow our code of conduct!