Abstract:Optimal page replacement is an important problem in efficient buffer management. The range of replacement strategies known in the literature varies from simple but efficient FIFO-based algorithms to more accurate but potentially costly methods tailored to specific data access patterns. The principal issue in adopting a pattern-specific replacement logic in a DB buffer manager is to guarantee non-degradation in general high-load regimes. In this paper, we propose a new family of page replacement algorithms for DB buffer manager which demonstrate a superior performance wrt competitors on custom data access patterns and imply a low computational overhead on TPC-C. We provide theoretical foundations and an extensive experimental study on the proposed algorithms which covers synthetic benchmarks and an implementation in an open-source DB kernel evaluated on TPC-C.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the optimization of buffer page replacement strategies in database management systems (DBMS). Specifically, the paper aims to design a new family of page replacement algorithms that can exhibit better performance than existing methods under custom data access patterns and maintain a low computational overhead in the TPC - C benchmark test. ### Problem Background In efficient buffer management, optimal page replacement is a key issue. Existing replacement strategies range from the simple FIFO (First - In - First - Out) algorithm to more precise but potentially costly methods customized for specific data access patterns. However, when adopting replacement logic for specific patterns, the main challenge is to ensure that performance does not degrade under high - load conditions. ### Main Contributions of the Paper 1. **Proposing a New Expert - type Page Replacement Algorithm (EEvA)**: Based on the expert - type algorithm framework, a new page replacement algorithm EEvA suitable for DB buffer managers is proposed, and three instantiated versions are provided: EEvA - Greedy, EEvA - Seq, and EEvA - T. 2. **Theoretical Basis and Experimental Verification**: By representing the changes in page replacement as a specific Markov decision process (MDP), the regret bounds of the EEvA algorithm are established. This shows that the EEvA algorithm has good performance guarantees in theory. 3. **Performance Evaluation**: By simulating different access patterns in a synthetic experimental environment, the superiority of the EEvA algorithm in terms of hit rate and latency is demonstrated. In addition, the implementation in an open - source database engine also proves the better performance of EEvA in the TPC - C benchmark test. ### Key Points of the Solution - **Light - weight Model**: The EEvA algorithm uses a light - weight model to capture data access patterns at different granularities (page, operator, table/query level). - **Online Learning Framework**: Through an online learning framework (such as the online mirror descent algorithm), EEvA can dynamically adjust the page replacement strategy to adapt to different data access patterns. - **Low Computational Overhead**: Despite the introduction of a complex model, EEvA still maintains a low computational overhead, ensuring efficient operation in high - load scenarios. ### Summary This paper solves the problem of insufficient performance of existing page replacement strategies under custom data access patterns by introducing the EEvA algorithm and its variants, and demonstrates its superiority and low computational overhead in the TPC - C benchmark test. This result provides a new and effective solution for buffer management in DBMS.

EEvA: Fast Expert-Based Algorithms for Buffer Page Replacement

Put an Elephant into a Fridge

PWLRU: a Buffer Replacement Algorithm for Flash-Based Database

Tl: A High Performance Buffer Replacement Strategy For Read-Write Splitting Web Applications

Operation-aware buffer management in flash-based systems.

Dynamic Access Distance Driven Cache Replacement

Efficient Cache Resource Aggregation Using Adaptive Multi-Level Exclusive Caching Policies

A Cache Replacement Policy Using Adaptive Insertion and Re-reference Prediction

Acar: An Adaptive Cost Aware Cache Replacement Approach For Flash Memory

AWRP: Adaptive Weight Ranking Policy for Improving Cache Performance

HFB: A page-block hybrid buffer manager for flash memory

From Cooperative Scans to Predictive Buffer Management

Buffer Allocation Algorithms for Embedded Real-Time Streaming File System

A Novel Cache Replacement Policy Via Dynamic Adaptive Insertion And Re-Reference Prediction

Managing Multiuser Database Buffers Using Data Mining Techniques

Optimal Bypass Monitor For High Performance Last-Level Caches

Paging with dynamic memory capacity

Performance Evaluations of Replacement Algorithms in Hierarchical Web Caching

Are Your Epochs Too Epic? Batch Free Can Be Harmful

Study of database buffer management approaches: Towards the development of a data mining based strategy

The Unexpected Efficiency of Bin Packing Algorithms for Dynamic Storage Allocation in the Wild: An Intellectual Abstract