Dynamic Server Allocation in Data Center Pods
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cloud data centers face challenges in optimizing resource utilization, network efficiency, and service stability due to non-blocking network topologies and resource sharing mechanisms, which lead to bandwidth underutilization and potential security issues.
Innovation Solution
A dynamic planning method for server resources in data centers that allocates servers using a fat-tree network topology, employing single pod and cross-pod reallocation procedures to optimize resource utilization and network efficiency, ensuring non-blocking network topologies and reducing interference between cloud services.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If current resource allocation mechanisms are used, then service diversity is supported, but resource utilization rate decreases due to network flow congestion and bandwidth underutilization
Solution Approach 1:
The data center network is segmented into multiple pods, with each pod containing multiple racks and servers. This segmentation allows independent resource allocation and management within each pod, enabling the system to accommodate diverse service requirements while optimizing resource utilization through localized resource pooling and allocation strategies
Solution Approach 2:
The patent introduces a hierarchical allocation dimension by organizing resources across multiple levels (data center → pod → rack → server). This multi-dimensional organization enables simultaneous support for service diversity at the data center level and optimized resource utilization at the pod and rack levels through structured resource pooling and allocation
2Productivity
If network resources are shared among cloud services, then resource efficiency improves, but network efficiency deteriorates due to mutual interference between transmission paths
Solution Approach 1:
Network resources are segmented by isolating different cloud services into separate pods or racks within pods. This physical and logical segmentation prevents transmission path interference between services while maintaining high resource efficiency through controlled sharing within isolated segments, eliminating the mutual interference that occurs in fully shared networks
Solution Approach 2:
Different regions of the network (pods and racks) are assigned different quality characteristics based on service requirements. High-performance services receive dedicated resources in isolated pods, while less demanding services share resources in other pods, creating local quality variations that optimize both resource efficiency and network efficiency simultaneously
3Productivity
If hybrid network resource sharing mechanisms are implemented, then resource utilization improves, but service stability deteriorates due to data security loopholes
Solution Approach 1:
The patent introduces isolation mechanisms as intermediaries between different cloud services, using pod and rack boundaries as protective barriers. These intermediaries enable resource utilization optimization through controlled sharing while maintaining service stability by preventing security loopholes and interference between services through hierarchical isolation layers
Data Source
AI summary
A dynamic planning method for server resources of a data center is provided. The data center includes multiple pods and each pod includes multiple racks configurable with multiple servers. In the method, a rack having a remaining space larger than or equal to d is searched from the pods and used for configuring d servers. If the rack cannot be configured with the d servers, a single pod reallocation is executed on one of the pods to empty a rack so that the remaining space of the rack is larger than or equal to d and configurable with the d servers. If there is no rack that can be emptied, a cross-pod reallocation is executed to empty corresponding server positions of corresponding racks in the pods so that the remaining space of the corresponding server positions is larger than or equal to d and configurable with the d servers.