Kademlia: XOR Distance and the DHT That Stuck
Maymounkov and Mazières: k-buckets, XOR metric, and why BitTorrent and Ethereum-style discovery still look like this paper.
Category
34 articles · 107 min of reading
Maymounkov and Mazières: k-buckets, XOR metric, and why BitTorrent and Ethereum-style discovery still look like this paper.
Castro and Liskov's PBFT: 3f+1 replicas, three-phase agreement, and when crash-fault Raft is the wrong threat model.
Fischer, Lynch, and Paterson 1985: why deterministic consensus cannot be both safe and live in a fully asynchronous network with one crash.
Lamport's 1978 CACM paper: happens-before, logical clocks, and why wall time is the wrong primitive for causality.
Sigelman et al. on Dapper: trace trees, sampling, and the production constraints that still shape OpenTelemetry and Zipkin.
Thomson, Diamond, Weng, Ren, Shao, and Abadi, SIGMOD 2012: agree on input order, then execute lock-free replicas in lockstep.
Thompson, Farley, Barker, and co. on the Disruptor: ring buffers, sequence barriers, and avoiding locks when microseconds are the product.
Cardwell et al. in ACM Queue: bottleneck bandwidth and RTT, why bufferbloat broke loss-based control, and what BBRv2 still argues about.
Malewicz et al. SIGMOD 2010: vertex-centric supersteps, combiners, and why graph algorithms stopped looking like giant MapReduce shuffles.
Langley et al. and the IETF QUIC RFCs: TLS 1.3 in the handshake, streams over UDP, and the middlebox tax HTTP/2 never escaped.
How Lakshman and Malik's SIGOPS 2010 Cassandra paper took Dynamo's consistent hashing, hinted handoff, and tunable quorum into a wide-column store.
Hunt, Konar, Junqueira, and Reed's USENIX ATC 2010 paper: a filesystem-shaped API, linearizable writes, and why watches beat homemade leader locks.
A working engineer's reading of Lamport's 2001 note: prepare/accept, majority quorums, and why Multi-Paxos still leaks into every consensus library.
Nakamoto 2008: proof-of-work as Sybil-resistant leader election, UTXOs, and the longest-chain rule without the hype cycle.
Stoica, Morris, Karger, Kaashoek, and Balakrishnan, SIGCOMM 2001: successor pointers, finger tables, and the DHT ring everyone still draws.
How 2PC actually runs, why a crashed coordinator freezes the world, and how Raft/Paxos-backed transaction coordinators changed the picture.
Cache-aside, read-through, write-through, write-behind — when each pattern fits, and the invalidation trade-offs nobody escapes.
How Bloom filters avoid disk reads in LSM trees and caches, how false positives behave, and when a cuckoo or ribbon filter is the better structure.
Brewer's conjecture, Gilbert and Lynch's proof, and why 'we picked AP' is usually a slogan rather than a design.
What the Raft paper actually guarantees, how leader election and log replication work, and the operational failure modes etcd and Consul still hit.
Bounded buffers, reactive pull-based demand, and log-based pull consumption: three ways streaming systems handle a slow consumer.
How the transactional outbox pattern makes updating a database and publishing an event atomic, without a distributed transaction.
Hash-based, range-based, and directory-based shard keys compared, and why resharding is the expensive part nobody plans for early.
CDNs now run edge compute, cache dynamic responses, and shield origin from thundering herds — far more than a cache for static files.