Distributed-Systems

Abstract visualization of distributed GPU clusters processing large language model inference requests

Distributed LLM Inference: Parallelism Strategies

Exploring the core parallelism strategies behind distributed LLM inference — tensor, pipeline, data, and expert parallelism — with real-world architecture patterns and practical deployment insights.

Parallelism strategies diagram showing concurrent task execution across multiple cores and nodes

Parallelism Strategies: Patterns, Pitfalls, and Production-Ready Approaches

Explore the core parallelism strategies—task, data, pipeline, and model parallelism—and learn how to choose the right pattern for your production workload, with concrete examples from Spark, Ray, Dask, and Kubernetes.

Abstract layered visualization of Apache Pulsar's hot and cold storage tiers

Inside Apache Pulsar: Architecting Tiered Storage for Petabyte-Scale Streaming Retention

How Apache Pulsar decouples hot and cold data using BookKeeper, segment stores, and pluggable offloaders to retain petabytes of streaming data affordably.

Diagram of a Raft cluster with one leader, two followers, and lease timing arrows.

Architecting Fault-Tolerant State Machines with etcd's Raft: Lease Semantics and Split-Brain Recovery

A practitioner’s deep dive into how etcd’s Raft implementation prevents split-brain through leader leases, quorum checks, and recovery protocols — with patterns for building fault-tolerant state machines on top.

A stylized cluster of three nodes with one leader highlighted, representing etcd's Raft consensus.

Inside etcd's Raft Implementation: Lease-Based Leader Election and Read Consistency

How etcd uses Raft with lease-based leader election, read-index protocol, and a backend MVCC store to deliver strict consistency for Kubernetes and other control planes.