FoundationDB high contention allocator

activesphere.com

62 points·wooola··9 comments

FoundationDB could really have been a big big thing had they not burnt bridges with their initial user set (i was one of them). It was a really cool database.

Care to provide more details ? (i don't understand why you got downvoted btw)

I think because when they sold to Apple they pulled their support and binaries.

Didn't Apple basically buy it as an acquihire and shut it down the next day?

FoundationDB had everything (launched or work in progress). SQL that was fast like NoSQL, No single point of failure, Scaling, Search, etc, etc... all built into one database. It was designed to remove the pain of managing many different components. For example, handling 'search' separately (via solr/lucene/cloudsearch) is a pain.

I'm not sure about the downvotes. I guess HN has lots of karma-police.

I was thinking about the « burning the bridge on initial users » part. But maybe you meant selling to Apple.

This post is a great illustration of the FoundationDB model. If you can design an algorithm that scales from a theoretical perspective, and map it onto the FoundationDB key-value model, writing the application code is as straightforward as your algorithm. You don't have to think about the details of storage, networking, and transactions very much.

FYI the FoundationDB community is doing its first community event in December: https://www.foundationdb.org/blog/announcing-foundationdb-su...

It took me a while reading the source code to convince myself that FoundationDB really doesn't use any prefix compression.

Seems like even with the directory encoding you'd be able to save a lot of storage space for keys by omitting the common prefixes.

Reply on news.ycombinator.com