Best Practices

To get the most out of your TigerGraph cluster, follow these architectural and development best practices.

1. Schema Design

• Primary IDs: Use strings for external IDs. TigerGraph handles the internal mapping.
• Attributes vs. Edges: If you need to filter on a value frequently, consider making it an edge to another vertex rather than just an attribute (for faster traversal).
• Data Types: Use the most compact data type possible (e.g., INT instead of STRING for numeric IDs).

2. GSQL Optimization

• Minimize Shuffle: Try to write queries that can be executed locally on each node before aggregating (Map-Reduce style).
• Limit Result Sets: Always use LIMIT in your SELECT statements during exploration to avoid huge network payloads.
• Accumulator Choice: Use SumAccum for counters and MaxAccum for finding best values. Avoid ListAccum on massive result sets as it can be memory-intensive.

3. Cluster Scaling

• Vertical Scaling: Increasing RAM and CPU on existing nodes. Best for handling complex queries on the same dataset.
• Horizontal Scaling: Adding more nodes. Best for increasing total storage capacity and handling more concurrent users (QPS).

4. Loading Performance

• Batch Size: Optimize your loading job batch sizes. Too small increases overhead; too large can cause memory spikes.
• Parallel Loading: Use multiple files and multiple loaders to saturate the system's I/O capacity.

[!IMPORTANT] Memory Allocation: Always reserve at least 20-30% of system RAM for the operating system and background processes to prevent GPE from crashing under high load.