Distributed Storage Decoupling Compute and Network Tasks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed storage systems face performance mismatches and constraints due to the evolution and deployment disorders between CPUs, storage drives, and networking entities, leading to inconsistent behavior and limited scalability, throughput, and increased costs.
Innovation Solution
The introduction of a 'FrontTop' device with a combination NIC that decouples computation, capacity, and bandwidth, allowing for a modular and scalable deployment of storage by merging access switch and NIC functionalities, and integrating storage SoC components with multiple storage drives, enabling efficient data management and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If computation and storage functionality are integrated in traditional distributed storage systems, then system simplicity is maintained, but performance mismatches and synchronization issues arise between CPU, storage drives, and networking entities
Solution Approach 1:
The patent segments the distributed storage system into separate functional components: storage nodes with storage drives, compute nodes with CPUs, and network entities with NICs. Each node type is optimized independently for its specific function, eliminating the performance mismatches that occur when all functions are integrated in a single node type.
Solution Approach 2:
The patent extracts computation functionality from storage nodes and places it in separate compute nodes. This extraction allows storage nodes to specialize in storage operations while compute nodes handle data processing tasks, improving overall system performance consistency.
2Quantity of substance
If storage nodes are equipped with high-capacity drives and high-throughput network, then storage capacity and bandwidth are improved, but synchronization challenges increase among CPU, storage, and network entities
Solution Approach 1:
The patent divides the system into specialized nodes where storage nodes focus on high-capacity storage and high-throughput network operations, while compute nodes handle CPU-intensive tasks. This segmentation reduces synchronization complexity by eliminating the need to coordinate all three entities within a single node.
Solution Approach 2:
The patent introduces a standardized interface layer between storage nodes and compute nodes that mediates data access and transfer operations. This intermediary layer abstracts the underlying hardware differences and provides consistent synchronization mechanisms across the distributed system.
3Ease of manufacture
If traditional distributed storage systems use integrated nodes, then deployment is simplified, but scalability and throughput are limited
Solution Approach 1:
The patent designs universal node types that can perform multiple functions through software configuration. Storage nodes can handle both storage operations and network communication, while compute nodes can perform various data processing tasks. This multi-functionality maintains deployment simplicity while enabling high throughput through specialized hardware.
Solution Approach 2:
The patent transitions from a single-dimension integrated node design to a multi-dimensional architecture where nodes are differentiated by their primary function (storage, compute, or network). This dimensional change allows the system to scale throughput by adding specialized nodes without complicating deployment procedures.
4Productivity
If more resources are allocated to each storage node, then individual node performance improves, but total cost of operation increases
Solution Approach 1:
The patent merges multiple functions into separate specialized nodes rather than concentrating all resources in each node. By combining storage nodes, compute nodes, and network entities into a coordinated distributed system, the architecture achieves high overall performance while allowing each node type to use resources efficiently for its specific purpose.
Data Source
AI summary
One embodiment provides a computer system. The computer system comprises: a plurality of storage devices; and a first component functioning both as a network interface card and as an access switch, wherein the first component is configured to manage connections to the plurality of storage devices. A respective storage device comprises: an Ethernet port coupled to the first component; at least one microprocessor; a plurality of PCIe lanes; and a plurality of storage drives with non-volatile memory.