# Improve query performance

Measure representative queries, choose indexes, and inspect query plans without relying on benchmark headlines.

Source: https://particledb.ai/guides/query-performance/

## Measure the query your application runs

Use representative record counts, value distributions, filters and concurrency. Record server version, hardware and configuration. Compare results before comparing timings.

## Read less data

Select the columns you need. Add selective filters and use a suitable index when the release supports it. Large offsets, wide records and high-cardinality grouping can change the cost substantially.

## Inspect a plan

```sql
EXPLAIN SELECT id, title FROM work_items WHERE id = 1;
```

This uses the sample table from [relational data](/guides/relational-data/). A plan is a diagnostic, not a latency promise. Use supported profiling tools to identify expensive work.

## Write in bounded batches

Compare individual writes with bounded batches. Preserve the same transaction and durability settings in both runs. Keep retries and memory use in the measurement.

## Separate examples and benchmarks

Documentation examples establish behavior. Benchmark results require a complete workload, configuration and repeated measurements. See [benchmark methodology](/benchmarks/overview/).

## Related

[Documentation home](/docs/) · [Troubleshooting](/guides/troubleshooting/)