tech_
how systems actually work — networking, AI, encryption, infrastructure
The 18,432-core difference between an H100 and a CPU is structural, not arbitrary
Running 1,000 matrix multiplications takes 15.6 seconds on an Intel Xeon and 0.3 seconds on an NVIDIA H100 because the hardware prioritizes throughput over latency.
What happens between kubectl apply and a running container
A kubectl apply command triggers a seven-step sequence across the API server, scheduler, and kubelet before a container starts.
Why a database sharded across 8 servers can be slower than 1
Sharding a database across eight servers introduces network latency to queries that would otherwise run locally on a single disk, making simple reads slower.
How HTTP/3 cuts connection latency by removing TCP handshakes
HTTP/3 moves transport from TCP to QUIC to eliminate round-trip delays. The change saves roughly 300 milliseconds on high-latency mobile connections.
How encryption keys actually work
Public-key cryptography relies on a mathematical trapdoor where creating a key is easy, but reversing it without the secret is computationally impossible.
Why a CDN has 3 cache layers, not 1
A single CDN edge layer creates a bottleneck; three layers distribute load by balancing latency, cost, and origin protection.
How the internet decides which path your packet takes — and why it sometimes goes wrong
BGP routing is trust-based, not optimized. A single misconfigured router announcement can pull half the internet's traffic into a black hole.
What happens in the 100ms between clicking a link and seeing the page
A browser click triggers five distinct network events, consuming roughly 100ms before the first byte arrives.