# Andy Pavlo

> ~1983– · Computer Scientist
>
> **Recorded contribution:** Database systems researcher; CMU; database history; systems education

## How to use this dossier

Read for a causal chain, not a hero story: inherited problem → contribution → mechanism → downstream capability → limit. Then close the page and complete the reconstruction exercise from memory.

## 1. Historical orientation

American computer scientist Andy Pavlo is a Carnegie Mellon professor specializing in database systems. His research includes transaction processing, benchmarking, in-memory systems, automatic database tuning, and query or workload analysis. He is also known for unusually open systems education through CMU's database courses and historical materials. The registry's broad 'database history; systems education' description is accurate but not one discrete invention. Pavlo's research treats database configuration as a feedback-control problem rather than folklore. Systems such as OtterTune learned relationships between workload measurements, configuration knobs, and performance from prior deployments, then proposed settings for a new database under observation.

## 2. The problem inherited

Database architecture decisions were often repeated without representative measurements, while increasingly complex systems demanded tuning expertise unavailable to many operators and students.

## 3. The central contribution

Pavlo advanced empirical database-system design and self-driving tuning research while publishing high-quality, implementation-centered database education.

## 4. Reconstruct the mechanism

1. Capture a workload with queries, data distribution, concurrency, and resource limits.
2. Instrument the DBMS to expose plans, counters, waits, and configuration state.
3. Use controlled benchmarks or learned models to predict a configuration or design intervention.
4. Validate on held-out workload phases and retain rollback when the environment changes.

## 5. What changed downstream

- His research influenced automatic tuning, benchmarking, and modern transaction-processing systems.
- Open CMU courses made database internals and implementation exercises accessible beyond enrolled students.
- This line of work advanced self-driving database research and made workload traces, reproducible benchmarking, and automated knob selection concrete engineering subjects rather than purely administrator intuition.

## 6. Attribution, limits, and uncertainty

- Pavlo's record is collaborative and broad; he should not be described as creator of database history or a single dominant DBMS.
- A learned tuner can overfit benchmarks, make unsafe changes, or optimize the wrong objective; human ownership and workload drift remain.
- A tuning policy can overfit one benchmark, violate latency or durability constraints not represented in its objective, or recommend unsafe changes when workload and hardware drift.

## 7. Reconstruction lab

Benchmark one query under five index or memory configurations. Separate warm-up from measurement, hold out one workload phase, and test whether the best training configuration generalizes. Reserve a workload shift for the final trial and require rollback criteria, showing whether the tuner learned a robust relationship or merely memorized the training regime.

## 8. Evidence trail

- [Andrew Pavlo](https://csd.cs.cmu.edu/people/faculty/andrew-pavlo) — Carnegie Mellon University
- [Andy Pavlo publications](https://www.cs.cmu.edu/~pavlo/publications.html) — Carnegie Mellon University
- [CMU Intro to Database Systems](https://15445.courses.cs.cmu.edu/) — Carnegie Mellon University

---

*Research checked 2026-08-09. Dates, roles, and claims about living people are historical snapshots. Linked sources remain the authority; this dossier is original instructional synthesis.*