Scalable Pooled NVMe Storage Box Architecture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current storage systems, particularly those utilizing NVMe drives, face challenges in scalability to meet the growing demands of data storage and processing needs, as they are limited by the bandwidth of SATA drives and lack efficient designs to manage data growth effectively.
Innovation Solution
A scalable cluster-architecture for a pooled-NVMe storage box is introduced, which comprises NVMe drives and switches, allowing the number of NVMe drives to be scaled up or down based on the number of nodes sharing the storage box, utilizing PCIe connectivity for high bandwidth and supporting dual or single ports to enable high availability clusters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If NVMe drives are pooled into a storage box to improve datacenter efficiencies, then power and cooling units can be shared among SSDs, but scalability of the NVMe storage box remains limited
Solution Approach 1:
The NVMe storage box is divided into modular components including multiple NVMe drives, switches, and network interfaces that can be independently configured and scaled. Each NVMe drive can be individually accessed by multiple nodes, allowing the storage capacity to be segmented and allocated dynamically based on demand without requiring complete system redesign.
Solution Approach 2:
The storage box design incorporates universal interfaces and protocols (NVMe over PCIe, switch fabric connections) that allow the same hardware platform to serve multiple functions and support various scaling scenarios. The system can accommodate different numbers of NVMe drives, switches, and network interfaces within a single standardized chassis, providing multi-functional adaptability.
2Quantity of substance
If the number of NVMe drives is increased to meet growing data storage demands, then storage capacity improves, but system complexity increases
Solution Approach 1:
Switches serve as intermediary devices that manage connections between multiple NVMe drives and multiple host nodes. The switch fabric abstracts the complexity of direct peer-to-peer connections, providing a standardized interface that simplifies routing and data management as the system scales from a few drives to many drives without proportionally increasing control complexity.
3Adaptability or versatility
If SATA drive bandwidth is used in conventional storage systems, then compatibility is maintained, but bandwidth is limited to a 600Gbps ceiling
Solution Approach 1:
The system transitions from SATA interface parameters (600Gbps ceiling) to NVMe over PCIe interface parameters, which offer significantly higher bandwidth capabilities. This parameter change in the interface standard enables speeds exceeding 600Gbps while maintaining backward compatibility through standard PCIe implementations and NVMe protocol support.
Data Source
Figure 1A
Figure 1B
Figure 1C~1D
AI summary
Various examples of the present technology provide a cluster-architecture to support a scalable pooled-NVMe storage box that can be shared among a scalable number of nodes. The scalable pooled-NVMe storage box comprises NVMe drives, one or more switches and one or more switch ports. The number of NVMe drives in the scalable scalablepooled-NVMe storage box can be scaled up or down based upon a number of nodes that need to share the scalable pooled-NVMe storage box.