Performance

Practical search performance guide

See practical search performance guide. The guide walks through a music search use case and gives a practical introduction to Vespa search performance.

Sizing and capacity planning

Sizing and capacity planning involves figuring out how many nodes are needed and what kind of hardware flavor best fits the use case:

• Sizing Vespa search: How to size a Vespa search cluster
• Caching in Vespa: How to enable caches in Vespa
• Attributes and memory usage: How attributes impact the memory footprint, find attribute memory usage
• Proton maintenance jobs: Impact on resource usage
• Coverage degradation: Timeout handling and Degraded Coverage

Benchmarking and tuning

Benchmarking is important both during sizing and for testing new features. What tools to use for benchmarking and how to tune system aspects of Vespa:

• Benchmarking Vespa: Test Vespa performance
• Search features and performance
• Feed performance
• Container Http performance testing using Gatling
• Container tuning: JVM, container, docproc
• vespa-fbench: Reference documentation
• HTTP/2: improve HTTP performance using HTTP/2

Profiling

Do a deep performance analysis - how to profile the application as well as Vespa:

• Profiling: Generic profiling tips.
• Valgrind: Run Vespa with Valgrind