RDMA Access Frequency Tracking via Client-Side Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Database systems face challenges in tracking access frequency when using direct memory access protocols, which bypass the database access mechanism, leading to lost statistics and inaccurate eviction decisions in buffer pool management.
Innovation Solution
Implement a method where clients collect and maintain access frequency data for records accessed via direct memory access, aggregating it until a threshold is reached, and then asynchronously report it to the database system to update internal data structures and implement eviction policies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If direct memory access protocol is used to access records from database memory, then access speed is improved, but access frequency tracking accuracy deteriorates because the database access mechanism is bypassed
Solution Approach 1:
The patent introduces a client-side intermediary component that intercepts RDMA access operations and collects access frequency statistics. This mediator sits between the RDMA protocol and the database memory, allowing fast direct memory access to continue while simultaneously tracking access patterns. The client maintains local data structures to record access frequencies and asynchronously reports this information to the database system, thus preserving both speed and measurement accuracy.
2Reliability
If access frequency data is collected and maintained for each record, then eviction decision accuracy is improved, but system complexity increases
Solution Approach 1:
The patent segments the access frequency tracking functionality into separate client-side data structures rather than requiring centralized tracking in the database system. Each client maintains its own access frequency counters and data structures locally, which reduces the complexity burden on the database system. The segmented approach allows the database system to receive pre-processed frequency information asynchronously, simplifying its eviction decision process while maintaining high accuracy through distributed tracking.
3Productivity
If access frequency data is asynchronously reported to the database system, then system performance is improved, but data freshness deteriorates
Solution Approach 1:
The patent implements periodic action by having clients asynchronously report access frequency data to the database system at intervals rather than continuously. This periodic reporting mechanism allows the system to maintain high performance by avoiding constant communication overhead, while the database system periodically updates its eviction decisions based on the accumulated frequency data. The approach balances performance and data freshness by finding an optimal reporting interval that satisfies both constraints.
Data Source
AI summary
A method includes the following steps. One or more records are accessed from a database memory bypassing a database access mechanism of a database system. Data representing access frequency of the one or more records are collected. The collected access frequency data for the one or more records are maintained. The access frequency data for the one or more records are aggregated until the access frequency reaches a threshold value. The aggregated access frequency data are asynchronously reported for the one or more records to the database system.