Cloud-Native Storage Management with Independent Capacity Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Enterprise datacenters struggle to manage and optimize the use of resources effectively, leading to over-subscription and underutilization of components, particularly in hyperscale cloud environments with data-intensive workloads like AI inferencing and analytics, resulting in inefficient use of investments.
Innovation Solution
A data management platform with disaggregated storage and compute resources, utilizing accelerator servers and compute servers, where storage capacity scales independently of compute, and resource allocation is managed through cache isolation and dynamic load balancing to meet stringent performance demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If additional servers are added to the datacenter to handle increasing data workloads, then storage capacity and processing power increase, but resource utilization becomes unbalanced with some components over-subscribed and others underutilized
Solution Approach 1:
The patent segments storage resources from compute resources, creating independent storage pools that can be dynamically allocated. Storage devices are disaggregated into standalone storage nodes that can serve multiple compute nodes, allowing storage capacity to scale independently without adding entire server configurations, thus improving resource utilization efficiency.
Solution Approach 2:
The patent creates universal storage pools that can serve multiple different workload types and compute nodes. The storage infrastructure is designed to be multi-functional, supporting various data-intensive workloads including AI inferencing and analytics through a common pooled storage system, eliminating the need for dedicated storage for each workload type.
2Speed
If storage and compute resources are tightly coupled in traditional servers, then data access speed is fast, but storage capacity cannot scale independently of compute power
Solution Approach 1:
The patent physically segments storage devices from compute servers, creating independent storage pools. Storage nodes are separated from compute nodes but connected through high-speed networking, allowing storage capacity to scale independently while maintaining fast data access through optimized network paths and caching mechanisms.
Solution Approach 2:
The patent introduces caching layers and buffer memory as intermediaries between storage pools and compute nodes. These intermediaries maintain high-speed data access by caching frequently accessed data locally at compute nodes while the bulk storage resides in independent pools, bridging the speed gap between coupled and disaggregated architectures.
3Adaptability or versatility
If resources are over-subscribed to handle peak workloads, then service coverage increases, but performance predictability decreases during high-utilization periods
Solution Approach 1:
The patent implements dynamic resource allocation where storage and compute resources can be flexibly assigned based on real-time workload demands. The system dynamically provisions storage capacity to compute nodes as needed, allowing over-subscription during low-utilization periods while maintaining performance predictability through on-demand resource availability during peak periods.
Solution Approach 2:
The patent incorporates monitoring and feedback mechanisms that track resource utilization and workload patterns. This feedback enables the system to adjust resource allocation dynamically, ensuring performance predictability by provisioning additional capacity when utilization thresholds are approached, while maintaining high service coverage through efficient resource sharing during lower-demand periods.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A data management platform that includes a compute server and a storage server is provided. The storage server manages a plurality of storage devices that are communicatively coupled to the storage server. The compute server and the storage server are communicatively coupled via a network. The plurality of storage devices that are managed by the storage server are disaggregated from the compute server to enable storage capacity of the plurality of storage devices to scale independent of the compute server.