Batch Deployment

• Batch processing generates predictions on a regular schedule and writes the results out to persistent storage to be consumed downstream
• Batch deployment is the most common deployment strategy
• Ideal for cases when:
  • Immediate predictions is not necessary
  • Predictions can be made in a batch fashion

Advantages

• Cheapest deployment method
• Ease of implementation
• Efficient per data point
• Can handle high volume of data

Disadvantages

• High latency
• Stale data
• Not suitable for dynamic or rapidly changing data.
• Not suitable for streaming data or real-time applications.

Liquid Clustering

<aside> 💡

Delta Lake’s optimization and performance features

</aside>

• Liquid clustering replaces table partitioning and ZORDER to simplify data layout decisions and optimize query performance