Linux & systems engineering
Kernel parameters, storage paths, scheduler behavior, systemd units, and the host-level details that decide whether a product stays predictable.
01 - Practice
Advanced engineering for production systems that have to stay up.
We design, tune, and recover the infrastructure under your product. Linux, HA clusters, high-load systems, cloud platform choices, observability, and the work that runs at 3 AM when the dashboards go red.
02 - Practice
What we work on.
Nine areas of work. We are not a 40-person agency with a long menu. These are the systems where experienced diagnosis changes the outcome.
High availability
Pacemaker and Corosync clusters, fencing, failover drills, split-brain prevention, and recovery procedures that have actually been exercised.
Performance & diagnostics
Latency budgets, profiling, packet captures, flame graphs, IRQ affinity, and the measurement discipline needed before changing production.
Cloud platform selection
AWS, GCP, Azure, DigitalOcean, bare metal, or hybrid: selection based on failure modes, cost shape, latency, operations maturity, and lock-in risk.
SPOF removal & system design
Architecture reviews that identify single points of failure, narrow blast radius, isolate dependencies, and turn failover behavior into an explicit design.
High-load architecture
Capacity planning, queue behavior, cache pressure, database hot paths, and load-shedding designs for systems that must stay boring under traffic.
Production recovery
Incident stabilization, root-cause analysis, rollback paths, dependency isolation, and the operational decisions that get a service back under control.
Observability & log systems
Monitoring, alerting, log management, metrics, traces, and dashboards built around failure detection, incident response, and operational decisions.
Automation & platform ops
Operational automation, deployment mechanics, observability, runbooks, and the small platform decisions that remove repeated human work.
03 - Engagements
Selected work, anonymized.
2025
Quote latency, tail cut from 4 ms to 0.6 ms
A trading platform was losing time in the host path after a kernel update. The NIC was not the bottleneck.
Pinned IRQs, corrected queue affinity, and removed a misleading autoscaling rule from the incident path.
2024
Cluster failover, 14-month clean run
A revenue-critical service had intermittent primary-node failures during maintenance windows.
Reworked health checks, fencing, and failover timing so traffic moved before user-visible failure.
2024
High-load API path made predictable
A customer-facing API had unpredictable tail latency whenever batch jobs and live traffic overlapped.
Separated queues, capped expensive work, documented overload behavior, and reduced manual intervention.
04 - On call
We answer the page.
Deep systems work is not a slide deck. We stay close to the logs, alerts, rollout mechanics, and failover paths that decide whether a system recovers cleanly.
- Median ack
- 2:14
- retainer incidents
- Observed SLA
- 99.95
- availability design target
- Stack depth
- 8
- kernel to protocol
05 - Writing
Field notes and post-mortems.
We publish what we can: logs, failure patterns, and operational notes from systems that made people think twice.
View all writing06 - Contact
Tell us what is broken.
A short email is enough. Include what changed, what failed, what you tried, and the first log line that made you stop.
To: [email protected] Subject: [P1] failover loop on cluster-02 PGP key: 9F4A 88E1 6C2D 41B7 0FA3 Status: https://status.msmsoft.com
Engagement intake
- Best fit
- Production systems, urgent diagnostics, architecture recovery.
- Not a fit
- Generic web builds, staff augmentation, commodity cloud migration.
- First reply
- Within one business day; same-hour triage for active incidents.