Reading
Research papers I've read and found worth sharing. Each entry includes my one-line takeaway and why I think it matters. I'm drawn to distributed systems, consensus, security, and applied cryptography.
The Ethereum whitepaper introduced the concept of a Turing-complete blockchain with smart contracts and a programmable decentralized state machine (EVM). It spawned the entire DeFi, NFT, and Web3 ecosystem. Critical reading for blockchain engineers building on XRPL, EVM-compatible chains, or any smart contract platform.
Borg is the container orchestration system that ran Google's workloads for over a decade and directly inspired Kubernetes. It introduced key concepts like resource quotas, priority scheduling, task grouping, and health checking at planetary scale. Understanding Borg is the foundation for understanding modern cloud-native infrastructure.
Cassandra combines the best of Dynamo (partitioning and replication) with Bigtable's data model into a highly available, leaderless NoSQL database. Now powering Apple, Netflix, and Discord at massive scale, Cassandra's design choices around tunable consistency are central to any NoSQL architecture discussion.
Jay Kreps' seminal essay on the log as the unifying abstraction for distributed systems — covering databases, stream processing, and data integration. It explains why the append-only log is the fundamental primitive behind databases, Kafka, CDC, and event sourcing. Required reading for any data/backend engineer.
The original Google paper. Introduced PageRank and the architecture of a large-scale web search engine including crawling, indexing, and ranking. It demonstrates how graph algorithms and large-scale distributed systems combine to extract knowledge from the web. The paper that started one of the most impactful technology companies in history.
Chord introduced the consistent hashing ring for distributed hash tables — the technique now foundational to load balancing, distributed caches (Memcached, Redis Cluster), and sharding strategies. Understanding Chord's O(log N) lookup via finger tables is key to designing scalable distributed systems.
The paper that invented relational databases. Codd's 1970 model laid the mathematical foundation for SQL, RDBMS, and 50+ years of database systems. Every engineer who has written a SQL query is building on this work. A Turing Award-winning contribution that fundamentally changed how we organize and query data.
The whitepaper that launched the blockchain era. Introduced proof-of-work consensus, the UTXO model, and a trustless peer-to-peer payment system. Whether building blockchain systems or not, every engineer working in fintech or distributed systems should understand the elegant engineering trade-offs in this 9-page paper.
Zanzibar is Google's authorization system serving billions of access-control checks per second for Drive, YouTube, Photos, and more. It introduced the relation-tuple model that spawned OpenFGA, SpiceDB, and Ory Keto. Foundational reading for anyone designing RBAC or ReBAC systems at scale.
Dapper defined distributed tracing as we know it. It inspired OpenTracing, Zipkin, Jaeger, and the entire observability ecosystem. For anyone building microservices, understanding how Dapper achieves low-overhead, always-on tracing across thousands of services is fundamental to production operations.
The paper that introduced Lamport clocks and the happens-before relation — the theoretical foundation of all distributed systems reasoning. It defines how we think about causality, event ordering, and consistency without a global clock. One of the most cited CS papers of all time.
The Transformer paper. Arguably the most influential ML paper of the decade — the architecture behind GPT, BERT, T5, and every modern LLM. The self-attention mechanism it introduced replaced recurrence and convolution, enabling parallelization and scaling that powered the generative AI revolution.
Lamport's accessible rewrite of the Paxos consensus algorithm. Paxos underlies Google Chubby, Zookeeper, and virtually every distributed coordination service. Though simpler than the original, it remains a conceptual cornerstone for understanding fault-tolerant consensus in distributed systems.
Raft was designed explicitly to be more understandable than Paxos while providing equivalent guarantees for consensus in distributed systems. It is now the consensus algorithm of choice in etcd, CockroachDB, TiKV, and many other production systems. Essential for understanding leader election and log replication.
Kafka's log-centric, append-only design for distributed messaging is one of the most impactful ideas in modern data engineering. It powers real-time data pipelines at LinkedIn, Uber, Netflix, and thousands of companies. Critical reading for anyone building event-driven or streaming architectures.
Spanner is the first system to provide globally-distributed ACID transactions with external consistency using TrueTime. It shattered the assumption that distributed systems cannot have strong consistency at global scale. The foundation for Cloud Spanner and modern globally-distributed NewSQL databases.
Bigtable introduced the sparse, distributed, persistent multi-dimensional sorted map model. It inspired HBase, Cassandra's data model, and wide-column NoSQL stores. Understanding Bigtable is key to grasping how Google Search, Maps, and Earth store and query petabytes of structured data efficiently.
Dynamo defined the practical approach to eventual consistency in distributed systems using consistent hashing, vector clocks, and sloppy quorums. It directly inspired Cassandra, Riak, and DynamoDB. A seminal paper on trading consistency for availability — central to the CAP theorem discussion.
GFS introduced the design principles behind distributed file systems at massive scale: chunk-based storage, single master, fault tolerance via replication. It directly inspired HDFS and underpins Google's infrastructure. A must-read for anyone building distributed storage systems.
The paper that defined the MapReduce programming model — a foundational abstraction for distributed data processing. It spawned Hadoop, Spark, and the entire big data ecosystem. Essential reading for understanding how to process petabyte-scale data across commodity hardware clusters.