I like to tinker and build. Growing up, it was Erector Sets, Lincoln Logs, and Legos. Now that I'm older I sometimes play with old muscle cars or tune firearms for target shooting. Overwhelmingly, I've always tinkered with code.
In days gone by, people would unambiguously refer to this tinkering as "hacking", and that's where the name of this blog comes from. In it, I explore different ideas about systems through my opinionated perspective. I am not one to play a game of research-as-marketing, and I'm definitely not a talking head blogging about the work of PhD students that I consider to be dogs pulling my sled. I am in the trenches, building systems that are (usually intricately) related to ideas I discuss.
These days, I'm a Ph.D. student in Computer Science at Cornell, specializing in distributed systems. I also serve as a distributed systems consultant.
Redupe is a command-line tool for forward error correction of bytestreams. Forward error correct makes small errors or bitflips easy to detect and correct. This post announces the 0.1.0 release of redupe.
read moreScrambler provides a deterministic mapping from 64-bit numbers onto 64-bit numbers. It guarantees there are no collisions in the output between distinct inputs and allows configuration of 2^64 different mappings. This post explains a couple use cases for scrambler by showing the problems it solves.
read moreGuacamole is a library for efficiently generating random data. Unlike most other random number generators, guacamole allows efficient indexing of the stream of data, so that the stream can be sliced and diced within tests or benchmarks.
read moreRelease announcement for signify ported from OpenBSD 6.1.
read moreNon-Volatile Memory (NVRAM) is rapidly becoming the focus for the next generation of data-backed applications. In this post we will look at some of the pitfalls that come from using NVRAM as a direct replacement for DRAM and propose some (hypothetical) solutions going forward.
read moreProof-of-Concept DB (pocdb) is a demonstration of how one could use Paxos to implement a strongly consistent key-value store. The code base is approximately 1,000 lines of code, and is intended to act as a teaching example that can be easily digested quickly.
read moreConsus is a new transactional key-value store that provides strong guarantees across data centers. Consus is a spiritual successor to HyperDex, another transactional key-value store. This post describes why I have switched my focus from working on HyperDex to working on Consus, what that means for HyperDex, and why I believe Consus to be the path forward.
read moreConsus is a geo-replicated transactional key-value store. This post introduces Consus, describes its open source release, and discusses the merits of releasing the code in advance of the publication. Finally, the post references other key-value stores related to Consus
read moreWhy does the world need yet another blog about systems?
read more