Persistent Memory Data Routing via Distributed Hash Structures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The transition to persistent memory (PMEM) as primary storage in information processing systems introduces challenges such as decentralized data access, changed data access semantics, accessibility issues for containers, multi-cloud domain data management, tenant partitioning, and data protection, as existing methods are not PMEM-aware and lack mechanisms for tracking redundant copies.
Innovation Solution
The implementation of routing structures and hash-based data structures to associate applications with PMEM locations, enabling per-tenant binding and global data reachability, using hash values to map applications to PMEM regions across servers, and managing PMEM storage as primary data storage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If centralized non-volatile primary storage is used for binding applications to persistent data, then data binding is achieved, but data access efficiency deteriorates and system complexity increases
Solution Approach 1:
The patent segments the centralized storage binding function into distributed routing structures at each host. Each host maintains local routing information that maps application identifiers to PMEM locations, eliminating the single point of contention in centralized storage while maintaining reliable binding through distributed consistency.
Solution Approach 2:
The patent introduces routing structures as intermediary components between applications and PMEM storage. These routing structures translate application identifiers into physical PMEM locations, enabling efficient indirect access without requiring applications to directly manage storage bindings, thus improving both reliability and access efficiency.
2Adaptability or versatility
If existing data access methods are used for PMEM, then compatibility with traditional storage is maintained, but PMEM-specific optimizations are lost and access semantics become unclear
Solution Approach 1:
The patent changes the access semantics parameters by treating PMEM as byte-addressable primary storage rather than block-based secondary storage. Applications access PMEM using load/store instructions with byte addresses, fundamentally changing the access pattern from traditional storage I/O to memory-like operations, which simplifies access semantics while enabling PMEM optimizations.
3Productivity
If containers are distributed across infrastructure components, then resource utilization improves, but binding containers to persistent data becomes complex and unreliable
Solution Approach 1:
The patent implements feedback mechanisms where routing structures continuously track and update the mapping between distributed containers and PMEM locations. When containers move or are recreated, the routing structures receive feedback about new locations and update bindings accordingly, maintaining reliable container-data associations despite dynamic distribution across infrastructure components.
Data Source
AI summary
In an information processing system comprising a set of computing devices wherein each computing device comprises a set of persistent memory modules resident in the computing device, a method generates at least one data structure that associates a given application program executing on the set of computing devices with at least one memory region of a given persistent memory module in a given one of the computing devices. The at least one data structure is utilized to route data between the given application program and the at least one memory region.