Skip to content
← ALL NOTES

Reading week: systems papers that are actually fun

Seven papers for people who claim they hate reading papers.

Once a year I take a week off client work and read. This year’s list skewed towards storage and clocks, which is what happens when you spend eighteen months getting merge logic wrong. Here are the seven that were worth the afternoon, in the order I would hand them to someone.

Start here

  1. “Time, Clocks, and the Ordering of Events in a Distributed System” — Lamport, 1978. Twelve pages, no prerequisites, and it quietly rearranges how you think about “before”.
  2. “The Google File System” — 2003. A masterclass in designing for the failures you actually have rather than the ones that sound impressive.
  3. “Dynamo: Amazon’s Highly Available Key-value Store” — 2007. Read it for the operational honesty; the sections on what went wrong are the best part.

The good papers are not the ones with the cleverest proof. They are the ones that tell you what the authors got wrong first.

If you liked those

  1. “Hybrid Logical Clocks” — Kulkarni et al., 2014. This is the one that ended up in bodega. Physical time you can compare, causality you can trust, and a design small enough to implement in an evening.
  2. “A Comprehensive Study of Convergent and Commutative Replicated Data Types” — Shapiro et al., 2011. Dense, but read the taxonomy and skip the proofs on a first pass.
  3. “The Log-Structured Merge-Tree” — O’Neil et al., 1996. Explains half the databases you use and most of their write-amplification complaints.
  4. “End-to-End Arguments in System Design” — Saltzer, Reed, Clark, 1984. Fifteen pages that will make you delete a caching layer.

How I actually read them

Three passes, and I only do all three for maybe one paper in four. First pass: title, abstract, section headings, conclusion — ten minutes, and it is fine to stop here. Second pass: the whole thing without the proofs, taking notes in a paper notebook so I cannot copy-paste. Third pass only if I intend to implement it, and then I write the smallest possible version before I trust that I understood it.

Papers we love and the ACM digital library between them cover everything on this list, most of it free. If a PDF is paywalled, the authors’ own pages very often are not.

— O.P., FORT GREENE, SEVEN FOR SEVEN