[Proposal] Enhancing Client-Server Data Transfer Protocols for Analytical Systems

[yjhjstz]

Proposers

No response

Proposal Status

Under Discussion

Abstract

N/A

Motivation

As data systems grow in scale and complexity, the performance of client-server communication protocols becomes a critical bottleneck, especially in analytical database workloads. The research project "Don't Hold My Data Hostage" has demonstrated significant inefficiencies in traditional client protocols used by popular databases (e.g., PostgreSQL, MariaDB, Oracle, Hive, MongoDB). Benchmarks show that existing ODBC/JDBC drivers are suboptimal for transferring large data volumes due to excessive serialization overhead and poor binary packing strategies.

Cloudberry, with its vision of high-performance, cloud-native data analytics, can benefit tremendously from protocol-level optimizations. We propose adopting and extending the benchmark framework from the Protocol-Benchmarks project to evaluate and redesign Cloudberry’s data transfer protocol to achieve optimal throughput and lower latency.

Implementation

• Implement the new protocol in the Cloudberry client and server modules.
• Ensure backwards compatibility with existing drivers where feasible.

Rollout/Adoption Plan

References

• Raasveldt, M., & Mühleisen, H. (2017). Don't Hold My Data Hostage: A Case for Client Protocol Redesign. VLDB.
• Protocol-Benchmarks GitHub Repository

Are you willing to submit a PR?

• Yes I am willing to submit a PR!