28th December 2024

"Systems ideas that sound good but almost never work"

I was sent this tweet today:

Systems ideas that sound good but almost never work:

- DSLs
- Live migrating process state
- Anomaly detection
- Control loops responding to system load
- Multi-master writes
- p2p cache sharing
- Hybrid parallelism
- Being clever vs over-provisioning

.. What else?

— martin_casado (@martin_casado) December 28, 2024

(Thanks Omar! I recommend following him, he has consistently surprising and fascinating opinions and things to share about computers, for a very broad definition of "computer".)

The replies also had some great examples (I recommend checking out the original tweet and reading the discussion though - it seems like tweeting about anything relating to "Systems" is still a reliable filter for good conversation on Twitter):

• Automatic transactions, also to KISS (Keep It Simple, Stupid) [tweet]
• Unifying batch and real time, Locality sensitive hashing [tweet]
• Row-level security [list]
• WASM/WASI/Function-as-a-service (specifically in the context of systems programming?) as opposed to using POSIX features like cgroups, existing POSIX virtualisation technologies like firecracker [tweet]
• multi-cloud [tweet]
• HATEOAS, especially resource discovery [tweet]

The tweet and its responses got me thinking:

• Most problems in computing have some "dumb" solutions that are simple, not optimally efficient (in space or time) but are applicable to a wide range of problems. Finding a specific, more efficient solution is satisfying as it justifies having learnt all the different solutions in college