Why this research matters
As ML systems mature, the biggest failures often come from data architecture, not model architecture. Schema drift, expensive migrations, and hidden query amplification in NoSQL systems can quietly degrade both performance and sustainability.
This work studies DaMoOps as an operational lens for managing data-model lifecycle decisions in production environments.
Core research question
How do we measure trade-offs between:
- short-term developer speed,
- long-term maintainability,
- infrastructure cost,
- and environmental footprint?
Framework
I evaluate NoSQL lifecycle decisions with four dimensions:
- Time: read/write latency, migration windows, retraining cycle impact.
- Money: storage growth, index overhead, compute and egress cost.
- Carbon: energy cost from repeated scans, over-indexing, and unnecessary replication.
- Obsolescence Risk: how quickly a schema becomes hard to evolve as product requirements change.
Method
- Benchmark embedded vs referenced schema patterns.
- Track workload behavior under real query mixes.
- Compare lifecycle outcomes at three checkpoints: launch, scale-up, and post-change adaptation.
- Map technical outcomes to decision policies for engineering teams.
Early findings
- Faster early delivery with denormalized structures can create expensive retraining and migration debt later.
- Index-heavy schemas reduce API latency but can increase background compute and carbon cost.
- Teams without lifecycle guardrails over-optimize for initial throughput and under-optimize for adaptation.
Next steps
- Publish benchmark results across MongoDB and Redis-backed workloads.
- Add a reproducible scoring template for Time x Money x Carbon decisions.
- Integrate lifecycle metrics into model governance checkpoints.