papershelf
Papers I keep going back to. New ones land in [queued];
annotations live in my notes, not here.
- Apache Hudi: The Data Lake PlatformVinoth Chandar, et al.VLDB '22
> Re-reading sections on the file layout and the timeline server while working on CDC ingestion at work.
#data-engineering#lakehouse
- Retrieval-Augmented Generation for Knowledge-Intensive NLP TasksPatrick Lewis, Ethan Perez, Aleksandra Piktus , et al.NeurIPS '20
> The original RAG paper. Useful to read after you've built a 'RAG system' and started suspecting what you built isn't quite what they described.
#ai#retrieval - In Search of an Understandable Consensus Algorithm (Raft)Diego Ongaro, John OusterhoutUSENIX ATC '14
> Worth re-reading after you've actually used etcd or Consul in production. The paper's clarity is the entire point.
#distributed-systems#consensus - Spanner: Google's Globally-Distributed DatabaseJames C. Corbett, et al.OSDI '12
> TrueTime is the headline, but the more interesting trick is what they're willing to spend on hardware to make timestamp ordering tractable.
#distributed-systems#databases - Dynamo: Amazon's Highly Available Key-value StoreGiuseppe DeCandia, Deniz Hastorun, Madan Jampani , et al.SOSP '07
> The paper that made consistent hashing, vector clocks, and sloppy quorums feel inevitable rather than exotic.
#distributed-systems#storage - MapReduce: Simplified Data Processing on Large ClustersJeffrey Dean, Sanjay GhemawatOSDI '04
> Read it for context, not because anyone writes MapReduce today. The framing of data locality and stragglers still shows up in every batch engine since.
#distributed-systems#data-engineering
- Attention Is All You NeedAshish Vaswani, et al.NeurIPS '17
> Re-reading carefully end-to-end. Wanted to ground a vague mental model in the actual equations.
#ai#transformers