Oasis: Hiding the Cost of Querying Parquet Files in the Datapath

2026-08-03

AI Summary

This work addresses the substantial CPU overhead incurred by Parquet decoding in cloud-native databases with disaggregated compute and storage, which undermines cost efficiency. It presents the first hardware acceleration approach that fully offloads Parquet decoding to a SmartNIC. By designing a custom hardware architecture and a complementary software abstraction layer, the solution seamlessly integrates into the DuckDB query engine and enables parallel execution of decoding alongside query processing within the network data path. This design effectively masks decoding latency, achieving near-doubling of system throughput in optimal scenarios while introducing minimal additional overhead.

Abstract

Cloud-native database systems disaggregate compute and storage resources to improve cost efficiency over traditional monolithic architectures through elasticity and resource pooling. Studies of production data warehouse workloads show that scans (including round trips to storage) account for roughly half of total query runtime. Data lakes and lakehouses amplify this bottleneck through per-query decoding of storage-optimized, compressed file formats such as Parquet. As storage and network bandwidth continue to outpace CPU cost-performance, the CPU cycles spent on decoding increasingly undermine the cloud's cost-efficiency promise. This has led to a wave of specialization across the stack with custom hardware at cloud-vendor scale at the extreme end. We build on this trend and present Oasis, a data-processing SmartNIC that offloads Parquet decoding into the network datapath as a custom hardware accelerator. Oasis features a hardware decoder architecture, software abstraction layer, and end-to-end integration with DuckDB. Our evaluation shows that Oasis hides the cost of Parquet decoding behind the network datapath with minimal overhead, overlapping the scan with the remainder of the query execution. In the best case, this almost doubles DuckDB query throughput.

Problem

• Parquet decoding
• cloud-native databases
• compute-storage disaggregation
• cost efficiency
• data lake

Innovation

• SmartNIC
• Parquet decoding
• hardware acceleration
• cloud-native database
• data lakehouse