Eliminating lock starvation in a high-concurrency service
before100%
after0%
How replacing a single global Redis lock with an append-only event log removed lock starvation and write races in a high-concurrency service.
Staff Software Architect · Austin, TX
Backend systems that hold up under real load.
I'm Mohammad Al-Khani. I design distributed systems, tune databases and build the platforms engineers ship on. 14 years of it.
lock.starvation100% → 0%lock starvation under peak traffic
api.latency−60%partner API response time
cloud.savings$2M+annual cloud savings
licensing.saved$10M+yearly client licensing removed
availability.slo99.999%availability SLOs
What I do
Five areas, each backed by shipped work.
/distributed-systemsDistributed systems
Locking, sharding and data models that hold up at peak traffic.
/database-performanceDatabase & query performance
SQL tuning, indexing, caching and memory engineering, measured with OpenTelemetry.
/platform-engineeringPlatform engineering & GitOps
Terraform, Atlantis, ArgoCD and Kubernetes paved paths with read-only cloud access.
/cloud-finopsCloud cost (FinOps)
Cost visibility and architecture changes that cut spend without cutting reliability.
/developer-experience-aiDevEx & AI governance
Reusable CI/CD, onboarding tooling and AGENTS.md review rules.
/case-studiesSee it in practice
Case studies with the problem, the approach and the measured result.
How I ship
Agile, two-week sprints. Planning and design come first; build, review and test overlap; the retro feeds the next sprint.
sprint(2w) · plan → design → build → test → release → retro↺ retro feeds the next sprint's plan
sprint2 weeks · 10 working days
1 · planrefine backlog, size, commit
2 · designdesign doc, ADR, review
3 · buildTDD, small PRs, CI
4 · test & reviewcode review, unit + functional
5 · releasefeature flags, canary, rollback
6 · retromeasure, learn, adjust
day 1day 6day 10
Case studies
Problem, approach, and the measured result.
Cutting cache memory 44%
before100%
after56%
Key-level memory profiling and a shared Redis client library cut cache memory 44%, peak utilization from 94% to 56%, and ended connection drops.
60% faster partner APIs
before1.0×
after0.4×
Finding slow SQL behind partner-facing APIs with OpenTelemetry tracing, then removing expensive joins, partitioning indexes and shrinking payloads.
About
I'm a Staff Software Architect and Backend Engineering Lead at General Motors, based in Austin, Texas. Over 14 years at GM, Life360, Upstart, CodeObjects and Oracle, I've decomposed monoliths, removed distributed locking bottlenecks, rebuilt caching layers and built the platforms other engineers depend on.
How I think
Measure first
A system is only as good as its observability and maintainability. I profile before I change anything.
Cost is an engineering metric
Savings come from changing architecture, not from cutting reliability.
Make the right way the easy way
Paved paths, reusable workflows and clear guardrails beat documentation nobody reads.
Raise the people around you
Mentoring, reviews and hiring are part of every senior engineer's job.