Outback: Fast and Communication-efficient Index for Key-Value Store on Disaggregated Memory

2025-02-13

AI Summary

In disaggregated memory architectures, RDMA-based key-value stores face a fundamental trade-off: one-sided operations avoid computation at memory nodes but incur multiple round trips, whereas two-sided operations achieve single-round-trip latency at the cost of imposing index lookup and other compute-intensive tasks on memory nodes—violating the principle of compute-memory separation. This paper proposes a dual-component decoupled indexing architecture that pioneers the separation of computation- and memory-intensive phases of dynamic minimal perfect hashing (DMPH): the computationally heavy DMPH construction is fully offloaded to compute nodes, while only a lightweight hash table resides on memory nodes for query processing. Integrated with an RDMA hybrid protocol combining one-sided and two-sided operations, our design achieves single-round-trip communication with zero computational overhead on memory nodes—the first such solution. Evaluation on a public-cloud prototype demonstrates throughput improvements of 1.06×–5.03× over state-of-the-art one-sided and two-sided baselines.

Technology Category

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

Application Category

• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs
• Security and Privacy: Data transparency and provenance
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Disaggregated memory systems achieve resource utilization efficiency and system scalability by distributing computation and memory resources into distinct pools of nodes. RDMA is an attractive solution to support high-throughput communication between different disaggregated resource pools. However, existing RDMA solutions face a dilemma: one-sided RDMA completely bypasses computation at memory nodes, but its communication takes multiple round trips; two-sided RDMA achieves one-round-trip communication but requires non-trivial computation for index lookups at memory nodes, which violates the principle of disaggregated memory. This work presents Outback, a novel indexing solution for key-value stores with a one-round-trip RDMA-based network that does not incur computation-heavy tasks at memory nodes. Outback is the first to utilize dynamic minimal perfect hashing and separates its index into two components: one memory-efficient and compute-heavy component at compute nodes and the other memory-heavy and compute-efficient component at memory nodes. We implement a prototype of Outback and evaluate its performance in a public cloud. The experimental results show that Outback achieves higher throughput than both the state-of-the-art one-sided RDMA and two-sided RDMA-based in-memory KVS by 1.06-5.03x, due to the unique strength of applying a separated perfect hashing index.

Problem

• Efficient indexing for disaggregated memory systems
• Minimizing communication round trips in RDMA
• Balancing compute and memory tasks in key-value stores

Innovation

• One-round-trip RDMA communication
• Dynamic minimal perfect hashing
• Separated index components