Sharding a Real-Time World with H3 Hexagonal Indexing
Why we chose Uber's H3 spatial index over quadtrees for partitioning a real-time multiplayer world, and what it costs.
Staff Software Engineer · Distributed Systems
Jeffrey Sylvan
I lead the design of backend platforms for high-concurrency, low-latency distributed systems.
Currently architecting a distributed real-time game engine targeting 10,000+ concurrent players with strict consistency guarantees and no time dilation. Previously Principal Engineer at CarGurus, where I led the decomposition of a monolith serving millions of daily users into domain-aligned microservices.
Featured case studies
Legend of the Black Skies
Scaling 10,000 Players: Distributed State in Real-Time Tactical Environments
H3 hexagonal spatial sharding, worker-per-region consistency, Redis/etcd/DynamoDB storage architecture. A distributed game engine designed to exceed EVE Online's concurrency without time dilation.
Distributed SystemsConsistency ModelsSpatial IndexingKubernetes
CogniCluster Engineering Practice
Architecture-Informed AI Delivery Pipeline
ADR-driven design decisions, spec-anchored AI coding workflows, Playwright and Storybook validation, GitHub Actions CI/CD with Terraform infrastructure.
CI/CDAI-Assisted DevelopmentTestingInfrastructure as Code
CarGurus
Monolith to Microservices at Scale
Led the decomposition of a monolithic Java application serving millions of daily users into domain-aligned microservices.
MicroservicesJavaDomain-Driven DesignMigration
VIDEO SERIES
Black Skies Architecture
A 7-episode technical deep-dive into building distributed systems for real-time multiplayer games. View all episodes →
1
Introduction & Architecture Overview
18 minProblem framing: why naive fan-out breaks at scale.
2
Massive Scale & SWIM Gossip
22 minNode failure detection and cluster membership at 10,000 players.
3
Designing Fan-Out for 10,000 Real-Time Players
25 minTile processor sharding, backpressure, and delivery guarantees.
How I build
I care about keeping implementation aligned with intent. In practice that means versioned design decisions, explicit context for AI coding tools, repeatable validation in Storybook and Playwright, CI/CD gates in GitHub Actions, and infrastructure defined in Terraform. The goal is not tool novelty for its own sake. It is making failures observable and requirements traceable.
Read about my engineering principlesRecent writing
Why etcd for Coordination When You Already Have Redis
The distinction between linearizable coordination and high-throughput hot state, and why Black Skies uses both.
Split-Brain Resolution with Epoch-Based Fencing Tokens
How CockroachDB's range lease model and Kleppmann's fencing tokens informed our approach to tile processor ownership.
Looking for a Staff or Senior Staff engineer?
Let's talk. I'm currently seeking Staff or Senior Staff IC roles in distributed systems, backend infrastructure, or platform engineering.
Recruiters: start with the about page, then skim the flagship case study. Download one-pager (PDF)