? Strategy 1: Multi-Reservation Architecture

Create:

• ETL reservation (5000 slots)

• BI reservation (2000 slots)

• Data Science reservation (1000 slots)

Assign projects accordingly.

Prevents:

• Dashboard users from blocking ETL

• ETL blocking ad-hoc analysts

? Strategy 2: Concurrency Caps

Limit:

• Max slots per query

• Max concurrent queries per group

Avoids single user monopolizing cluster.

? Strategy 3: Time-Based Isolation

Schedule heavy ETL:

• Nighttime

• Low BI usage hours

? Strategy 4: Autoscaling Buffer

Set:

• Baseline slots for average

• Autoscaling for peaks

Avoid overcommitting expensive idle slots.

? Strategy 5: Query Tiering

Define:

Use separate reservations.

? Mental Model at 100TB+ Scale

Think of BigQuery as:

Storage → Parallel Scan
Shuffle → Distributed Hash + Memory Pressure
Slots → CPU + Memory Unit
Skew → Single worker bottleneck
Reservations → Compute governance
BI Engine → Memory accelerator

? At True Enterprise Scale (500TB+)

What really matters:

• Shuffle minimization

• Skew elimination

• Pre-aggregation

• Workload isolation

• Slot autoscaling discipline

• Monitoring slot_ms daily

• Breaking monolithic queries