Stuff I read last week

About: A collection of links and short commentary, published weekly. In theory the inclusion criteria are for it to be something I read last week (duh) and was either particularly interesting or something I might want to refer to later. Subscribe with RSS or see my main blog for long-form writing.

2017-09-26

Proximity of "house address twins"
It's possible for multiple houses (e.g. in different cities) to have the same street name and number. What's the closest pair of such "twins" in the UK?

I truly appreciate the dedication that went into researching this bit of trivia.
What’s So Bad About POSIX I/O?
Why POSIX filesystem semantics aren't a good fit for large scale systems. (The funny thing is that it never occurred to me that anyone would use POSIX APIs in a modern distributed context. But apparently in supercomputing they do).
Mison: A Fast JSON Parser for Data Analytics Execute queries on a JSON file without parsing the file proper. Do a quick SIMD-based pass that gets the field/value locations. Do a full parse of a small sample of the file. Use the sample to predict the structural locations of the fields the user is interested in. Combine these bits of information to speculatively find the physical locations of the relevant fields.
COST in the land of databases > In the past two years the database community has gotten substantially worse at doing graph processing with SQL. I wouldn't recommend it, really. They've gone from a 16-core machine being only 10x slower than a single-threaded for-loop, to being up to 1000x slower but only on graphs up to 500MB in size.
Rendezvous Hashing: My Baseline “Consistent” Distribution Method > Regardless of the reason for consistent hashing’s popularity, I feel the go-to technique should instead be rendezvous hashing. Its basic form is simple enough to remember without really trying (one of those desert island algorithms), it is more memory efficient than consistent hashing in practice, and its downside–a simple implementation assigns a location in time linear in the number of hosts–is not a problem for small deployments, or even medium (a couple racks) scale ones if you actually think about failure domains.
Building multiplayer games with Socket.io and HTML5 Canvas The key insight here is speculation that the reason all the games like agar.io control like crap (small acceleration, tons of momentum) is that it allows for hiding all the networking flakiness.

◀ Earlier | Index | Later ▶