AI Infrastructure
KV cache
Retained attention state connects model geometry to allocation strategy and serving capacity.
- explains KV cache memory per token (article)
- demonstrates KV Cache Visualizer (experiment)
- source PagedAttention (paper)
08 / Connected ideas
Follow the connections.
Start with a concept, then find the article, experiment, or primary source that explains it from another angle.
Article / explainsExperiment / demonstratesProject brief / investigatesBenchmark / measuresPaper or documentation / source
Distributed Systems
Consistent hashing
A placement mechanism for changing membership. Related to balance and key movement, not a replication protocol by itself.
- explains Consistent hashing and key movement (article)
- demonstrates Hash ring experiment (experiment)
- measures Hash placement microbenchmark (benchmark)
- investigates Distributed Cache brief (project)
Distributed Systems
Replication & quorum
Copies create choices about durability, staleness, and availability. Set intersection is only one part of a consistency argument.
- demonstrates Replication & quorum experiment (experiment)
- source Dynamo paper (paper)
- investigates Distributed Cache brief (project)
Linux Internals
Memory locality
CPU placement, page placement, and I/O attachment describe different parts of a topology. Keeping them aligned is a question to measure.
- demonstrates NUMA Topology Explorer (experiment)
- source Kernel memory policy documentation (paper)
Performance Engineering
Tail latency
Distributions connect individual observations to service behavior. Sample size and measurement method determine what is safe to conclude.
- explains Latency percentiles and the limits of averages (article)
- demonstrates Latency Analyzer (experiment)
- measures Local measurement dashboard (benchmark)
- investigates Observability Agent brief (project)
Systems Design
Bounded work
Queues, admission control, and cancellation turn overload into an explicit policy rather than an accidental failure mode.
- demonstrates Token bucket experiment (experiment)
- investigates Async RPC Framework brief (project)
- investigates C++ Thread Pool brief (project)
Storage
Cache eviction
Replacement policy encodes a prediction about future reuse. LRU and FIFO make different predictions even for the same access trace.
- demonstrates Cache Simulator (experiment)
- investigates Mini Storage Engine brief (project)
No matching concepts. Try memory, cache, Linux, or latency.
Each connection states its purpose: explain a mechanism, demonstrate an idea, or measure an implementation. Project briefs remain proposals, not completed code.