Performance
Browse all articles, tutorials, and guides about Performance
Posts
ndots:5 in Kubernetes: Why One DNS Lookup Turns Into Ten Queries
A normal external hostname can make eight or ten DNS requests from a Kubernetes Pod before it succeeds. Trace the resolver rule behind that fan-out, measure it, and choose fixes that do not break cross-namespace Service names.
Docker Build Cache on Ephemeral CI Runners: inline, registry and local, Measured
A fresh CI runner starts with no BuildKit cache. We rebuilt a real multi-stage Node service on a new builder every time, with no cache, inline, registry min and max, and a local cache directory, and measured cold builds, a one-line change, a dependency bump, the export cost and the cache size.
Terraform 1.17's -minimal-refresh: 202 API Calls Down to Zero, and the Drift Goes Quiet Too
Terraform 1.17 can skip the refresh for every resource you did not change. We measured it on 202 real DigitalOcean resources: API calls went from 202 to 0 and plan time halved, and then we broke three records behind its back to see what it stops noticing.
Where Your Code Runs When There Is No Region
The edge compatibility tax people still warn about has mostly been paid off. What is left is stranger and more interesting: you can keep your Node imports, but you cannot keep your process. Two measured demos of what that actually costs.
Serverless Killed Your Connection Pool
A connection pool only works because a process outlives the request. Take the process away and it has nothing to amortise across. Measured against a real Postgres: what a connection costs cold and warm, why the pooler barely touches latency, and the benchmark that makes pooling look pointless.
Stop Paginating With OFFSET: Keyset Pagination and the Deep-Page Cliff
LIMIT/OFFSET is the reflexive way to paginate, and it is fine on page one. But OFFSET makes the database generate and throw away every row before the page you want, so cost grows with page depth until deep pages fall off a cliff. Keyset pagination makes every page the same speed regardless of depth. Here is why OFFSET is slow, the fix, and the one tradeoff.
Stop Using Random UUIDs as Primary Keys: uuidv7() Lands in PostgreSQL 18
Random UUIDv4 primary keys quietly wreck insert speed and bloat indexes on large tables. PostgreSQL 18 ships a native time-ordered uuidv7() that keeps the upsides of UUIDs without the B-tree penalty. Here are the numbers and how to adopt it.
How Does It Work So Fast? The Engineering Behind Instant UI Responses
Credit card validation, username checks, autocomplete, URL shorteners - they all feel instant. Here is what is actually happening under the hood in each case.
Docker Image Optimization: Best Practices for Smaller, Faster Images
Learn proven strategies to optimize Docker images: multi-stage builds, layer caching, base image selection, and security hardening. Reduce image size by up to 90% while improving build times and security.
Increasing the Maximum Number of TCP/IP Connections in Linux and Monitoring Network Usage
Find the limit you actually hit, then raise it: file descriptors, the ephemeral port range, listen backlogs, TIME_WAIT and connection tracking. Then watch bandwidth per process, per connection and per interface with nethogs, iftop, nload and ss.
UDP vs TCP: How Much Faster Is UDP Really?
How much faster UDP really is, measured on localhost, where the difference comes from, and when it is appropriate to use UDP instead of TCP: DNS, calls, games, VPNs and QUIC.
How to Profile C++ Code Performance on Linux
Learn how to use profiling tools like gprof, Valgrind, and perf to analyze C++ application performance and identify bottlenecks on Linux systems.
How to Simulate a Slow Internet Connection for Testing
Throttle and break the network on purpose: tc netem on Linux, Network Link Conditioner on macOS, Clumsy on Windows, browser DevTools, Toxiproxy for a single service, and three ways to create a connection timeout.