Choose File Storage Tiers Using Access Predictions
Company: Apple
Role: Machine Learning Engineer
Category: ML System Design
Difficulty: medium
Interview Round: Technical Screen
Design a method for choosing the storage tier for a user's file when tiers have different storage and access costs. The interview frames future file access as a classification problem: predict whether a file will be accessed and use that prediction to help select a tier. Frequently accessed and archival Amazon S3 storage classes are examples, not a prescribed implementation.
### What a Strong Answer Covers
- A clear prediction horizon, label definition, and decision objective.
- Features available when the tier decision is made and prevention of future-information leakage.
- A cost-aware tier choice that also respects retrieval requirements.
- Calibration and evaluation of access probabilities and the resulting storage policy.
- Cold files, new files, changing access patterns, and the cost of moving between tiers.
### Follow-up Questions
- Why might an accurate access/no-access classifier still produce an expensive storage policy?
- When is the probability of at least one access insufficient for deciding between tiers?
Overview: Choose file storage tiers using calibrated access probabilities, retrieval requirements, expected costs, leakage-free evaluation, and movement-aware policies.
Read the full Apple Machine Learning Engineer interview experience this question came from