Guidelines for Building Indexes on Partially Cache-Coherent CXL Shared Memory

2025-11-09

AI Summary

Existing single-node indexes fail on Partial Cache Coherence (PCC) platforms due to the absence of global cache coherence guarantees. Method: This paper proposes an index reconstruction framework integrating SP (Shared-Persistent) and P³ principles—namely, in-place updates, replicated shared variables, and speculative reads—to systematically address concurrency control challenges between sync-data and protected-data under PCC. Leveraging CXL-shared memory modeling, synchronization optimization, and speculative execution, the framework ensures verifiably correct index consistency. Contribution/Results: Experimental evaluation demonstrates up to 16× index performance improvement on PCC platforms, significantly outperforming message-passing and decoupled-memory baselines (by 16× and 19×, respectively). To our knowledge, this is the first efficient, coherent, and deployable indexing paradigm tailored for PCC environments.

Technology Category

• Search and Optimization: Distributed Search
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Data Mining & Knowledge Management: Scalability, Parallel & Distributed Systems

Application Category

• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web crawling and indexing

Abstract

The emph{Partial Cache-Coherence (PCC)} model maintains hardware cache coherence only within subsets of cores, enabling large-scale memory sharing with emerging memory interconnect technologies like Compute Express Link (CXL). However, PCC's relaxation of global cache coherence compromises the correctness of existing single-machine software. This paper focuses on building consistent and efficient indexes on PCC platforms. We present that existing indexes designed for cache-coherent platforms can be made consistent on PCC platforms following SP guidelines, i.e., we identify emph{sync-data} and emph{protected-data} according to the index's concurrency control mechanisms, and synchronize them accordingly. However, conversion with SP guidelines introduces performance overhead. To mitigate the overhead, we identify several unique performance bottlenecks on PCC platforms, and propose P$^3$ guidelines (i.e., using Out-of-underline{P}lace update, Reunderline{P}licated shared variable, Sunderline{P}eculative Reading) to improve the efficiency of converted indexes on PCC platforms. With SP and P$^3$ guidelines, we convert and optimize two indexes (CLevelHash and BwTree) for PCC platforms. Evaluation shows that converted indexes'throughput improves up to 16$ imes$ following P$^3$ guidelines, and the optimized indexes outperform their message-passing-based and disaggregated-memory-based counterparts by up to 16$ imes$ and 19$ imes$.

Problem

• Maintaining index consistency under partial cache-coherence model
• Reducing performance overhead in converted indexes on PCC platforms
• Optimizing index efficiency using novel P³ design guidelines

Innovation

• Using sync-data and protected-data for cache coherence
• Applying Out-of-Place update and Replicated variables
• Implementing Speculative Reading to enhance performance