Last year at Summit, we announced the public launch of the Snowflake Performance Index (SPI), an aggregate index for measuring real-world improvements in Snowflake performance experienced by customers over time. 

At Snowflake, our product philosophy is focused on continuously enhancing performance by refining the core engine. We deliver these performance improvements through weekly releases, integrating them into your daily operations to boost performance effortlessly and at no additional cost, with no migrations, upgrades or manual work required. This strategy is designed to optimize your price-for-performance with Snowflake. Then, with our consumption-based pricing model, these performance improvements can translate to cost savings for customers. 

Our commitment is to decrease your cost of running workloads in Snowflake over time.

Snowflake Performance Index Results

The SPI has now improved by 27% since we started tracking the index in August 2022. Over the last 12 months, it has improved by 12%. And since the SPI is calculated on real-world stable and recurring customer workloads, it allows us to compare improvements on specific customer workloads over time.  

Unlike other vendors, who may use fictitious or synthetic benchmarks to make performance comparisons, Snowflake uses real customer data on production workloads to measure the performance enhancements we make — a testament to Snowflake’s core philosophy: Our success is intrinsically tied to our customers’ success.  

Performance improvements tracked by the SPI

Over the last several months we’ve made a number of significant improvements. And best of all, many of these improvements occur automatically, with no knobs to tune or actions to take.

We continue to invest in making the compiler faster and more efficient; making the optimizer more intelligent; improving the core query execution performance; and making data ingestion into Snowflake even faster. For example:

• We improved the ingest performance of both JSON and Parquet files with case-insensitive data up to 25%. 
• We improved throughput between the nodes in a warehouse, by introducing faster intra-execution node communication and better network compression, and made improvements to aggregation placement, improving performance for all queries.  
• We identified and optimized query performance for common query patterns, such as improving memory management with holistic broadcast join decisions, which quickens the execution time for queries with deep right join trees. 
• We also reduced the cost of maintaining materialized view by improving the utilization of service resources. 
• And we continue to make the optimizer more intelligent — to choose the best possible optimizations, such as introducing more granular selectivity estimations, which helps Snowflake make better decisions on join orders.  

At Summit 2024 specifically, we’re proud to have announced the following performance optimizations:

• Top-K pruning V2 (generally available): Snowflake is enabling precise pruning with micropartitions at query runtime for SELECT statements that have ORDER and LIMIT clauses. With Top-K pruning V2, Snowflake stops scanning when it determines that none of the remaining rows can be in a result set that consists of K records. This pruning algorithm is automatically applied to ORDER and LIMIT queries, at no additional cost, to provide users with the best performance and experience. In fact, TopK v2 further improves performance by 12.5% on average. 
• DML History and Account Usage Views (generally available soon): This helps customers analyze the cost and benefit of clustering and search optimization to improve pruning, which provides them more transparency in understanding how efficiently queries run. 
• Improvements to the Query Acceleration Service (QAS): Now, 6% more queries are eligible for QAS with the addition of INSERT. 
• Continued clustering performance improvements (generally available): These would reduce costs for customers by 10%*.

You can read more about the latest improvements in our performance release notes.

The SPI highlights our commitment to continuously improve economics for customers, and it provides transparency on the quantitative impact of platform performance improvements on customers’ real production workloads over time. And of course, it allows us to measure and improve the performance impact of new features, enhancements and compute options for our customers, true to Snowflake’s No. 1 value: putting customers first. 

Visit the SPI website to learn more. 

*Based on a limited rollout from 2/12 – 3/1  of ~37% worth of reclustering credits 

The post Snowflake Improves Performance by 27%, According to the Snowflake Performance Index appeared first on Snowflake.