Disaggregated Storage Allocation Areas for Cross-Node Space Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing distributed storage systems suffer from sub-optimal infrastructure architectures that lead to inefficient utilization of compute, network, and storage resources, resulting in storage silos and limited flexibility in disk space allocation across nodes.
Innovation Solution
Implementing a disaggregated storage architecture with dynamically extensible file systems (DEFS) that allow nodes to access and own allocation areas (AAs) of a shared storage pod, enabling flexible space balancing and independent scaling of performance and capacity without internode communication overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If each node is associated with a dedicated pool of storage space (node-level aggregate), then storage space is isolated and manageable at node level, but storage silos are created and disk space cannot be used fluidly across nodes
Solution Approach 1:
The storage system is segmented into multiple allocation areas (AAs), each independently owned by different DEFS. This segmentation allows flexible reassignment of storage units (AAs) between nodes while maintaining organized management, resolving the contradiction between node-level manageability and cross-node flexibility.
Solution Approach 2:
The storage architecture transitions from static node-level aggregates to dynamic allocation areas that can be reassigned between DEFS based on space balancing requests. The quality attribute and dynamic ownership transfer enable storage space to adapt flexibly across nodes while maintaining operational control.
2Productivity
If storage space is partitioned into multiple allocation areas with different quality attributes, then space balancing can be optimized, but metadata management complexity increases
Solution Approach 1:
Different allocation areas are assigned different quality attributes (e.g., performance, capacity, reliability) based on their characteristics and the requirements of the DEFS that own them. This local quality differentiation enables optimized space balancing while keeping metadata manageable through attribute-based classification.
Solution Approach 2:
The system uses quality attributes as parameters to characterize allocation areas and guide space balancing decisions. By changing and adjusting these quality parameters based on system needs and AA characteristics, the system optimizes space balancing efficiency while maintaining structured metadata management.
3Adaptability or versatility
If disaggregated storage architecture is implemented with DEFS, then storage space can be used fluidly across nodes, but internode communication overhead increases
Solution Approach 1:
Each DEFS autonomously manages its own allocation areas and makes space balancing requests based on its own space needs. The donor DEFS independently identifies and transfers AAs to recipient DEFS without requiring constant coordination or communication overhead, enabling storage fluidity while minimizing internode communication.
Data Source
AI summary
Systems and methods for transferring ownership of allocation areas (AA) and associated metadata information to facilitate space balancing within a distributed storage system that makes use of disaggregated storage are provided. In various examples described herein, the unit of storage space assignment used to associate blocks of storage space within a storage pod with a given dynamically extensible file system (DEFS) is an AA, representing a large chunk of one or more gigabytes (GB). The use of AAs described herein allows disk space to be used more elastically across individual nodes of a storage cluster. When partial AAs are transferred among DEFSs, the transfer of associated metadata information may make use of various optimizations to minimize the amount of data copying performed. For example, copying physical volume block numbers (PVBNs) directly or indirectly indicative of locations of metafile data is faster than copying the underlying metafile data.