Performance testing¶

Understand your use case¶

Testing for the outcome you are looking for is very important. Understanding how hybrid tables will augment your architecture is important when designing your tests.

Design your test scenario:

• Do you require a high volume of UPDATE, INSERT, or DELETE statements?

• Does your application need fast access to indexed data?

• Do you have batch jobs you would like to run more often without impacting SELECT performance?

• What do you want to measure during the test?

Select a test framework¶

Performance testing frameworks are ubiquitous in software development. Most customers have testing frameworks that are already in place and can be used to test hybrid tables. Regardless of the test framework you select, it needs to be able to:

• Authenticate with Snowflake using shared key authentication

• Support multi-threaded query execution

• Issue queries as prepared statements, binding variables as needed

• Create a mix of INSERT, UPDATE, DELETE, and SELECT queries

Ideally, your framework will track query execution time for each request in each thread to calculate:

• Total query throughput

• Min, max, average, and standard deviation of response time

• Total bytes received per query

Execute the test¶

The hybrid tables query optimizer takes some time to “warm up” and establish a steady-state latency. This warm-up period can vary based on the amount of data, the number of indexes, and the complexity of the query. For most test cases, a warm-up period of 1-2 minutes is sufficient. Longer warm-up periods may be required.

This is a typical performance test result for random queries on a single hybrid table. Note that the performance improves over time and achieves a steady state after a few seconds: