Engineering
How we eliminated cold starts: 18 months of runtime engineering
Cold starts were killing us. Developers expected sub-millisecond response times, but our V8 initialization was taking 50ms. Here's how we obsessed over every microsecond, reworked our entire execution pipeline, and brought cold starts down from 50ms to 1.8ms p99.
This is the story of isolate pooling, garbage collection tuning, and the midnight incident that changed our entire approach to performance.
Marcus Wei
Apr 12, 2026
Read full article →
⚡
Architecture
Edge KV consistency: why we chose eventual
Strong consistency sounds good, but it would destroy global performance. We chose eventual consistency with conflict-free replicated data types. Here's why.
Julia Nakamura
Mar 28, 2026
Read article →
Tutorial
Compiling Rust to WASM at the edge: the full pipeline
Step-by-step walkthrough of our Rust-to-WASM compilation pipeline. Learn how we optimize for size, startup time, and security in edge environments.
Raj Patel
Mar 15, 2026
Read article →
Performance
Streaming AI responses through 320 regions without buffering
LLM responses need to stream immediately. We built a global streaming infrastructure with <10ms latency overhead. The architecture and benchmarks.
Emma Larsson
Feb 29, 2026
Read article →
Architecture
Why we ditched containers for V8 isolates
Containers are heavyweight. V8 isolates gave us 10x faster startup, 50x less memory overhead, and multi-tenant execution on single machines. The trade-offs explained.
Marcus Wei
Feb 14, 2026
Read article →
Engineering
Edge D1: SQLite at global scale
Running SQLite at the edge requires rethinking consistency, conflict resolution, and disaster recovery. Here's how we made it production-ready.
Julia Nakamura
Jan 31, 2026
Read article →
Performance
Building observability for serverless without breaking the bank
Distributed tracing across 320 regions is expensive. Our observability stack reduces costs by 70% while maintaining full debugging capabilities.
Klaus Bergmann
Jan 10, 2026
Read article →