Cluster Server Provisioning Based on Throughput and Load Capacity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers often operate with an overprovisioned cluster due to peak load assumptions, leading to increased operational costs and energy wastage, as most clusters do not operate at peak load conditions for majority of their time.
Innovation Solution
A method to optimize the number of application and backend servers in a cluster by determining the required number based on storage capacity, round trip time, network interface card bandwidth, and load bearing capacity, with the management server provisioning additional servers if actual throughput is less than the maximum throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of servers is increased to handle peak load conditions, then the cluster can bear maximum load, but operational costs and energy consumption increase during non-peak periods
Solution Approach 1:
The patent implements dynamic server provisioning that adjusts the number of active servers based on real-time load conditions. The management server continuously monitors cluster utilization and provisions or de-provisions servers dynamically, transitioning from static peak-load-based provisioning to adaptive provisioning that matches actual demand, thereby reducing energy waste during low-utilization periods while maintaining reliability during peak periods
Solution Approach 2:
The system changes the operational parameters of the cluster by adjusting the number of active servers based on monitored performance metrics such as throughput, utilization percentage, and load conditions. This parameter adjustment allows the cluster to operate efficiently at varying capacity levels rather than maintaining fixed peak-capacity configuration, resolving the contradiction between maintaining high load-bearing capacity and reducing energy consumption
2Productivity
If more servers are provisioned to ensure adequate throughput, then cluster performance is maintained, but boot time and migration difficulty increase
Solution Approach 1:
The management server performs preliminary assessment of cluster requirements by analyzing current throughput, utilization metrics, and workload characteristics before provisioning new servers. This preliminary evaluation ensures that servers are only added when genuinely necessary, avoiding unnecessary boot times and migration operations while maintaining adequate throughput capacity
Solution Approach 2:
The system implements feedback mechanisms that continuously monitor cluster performance metrics including throughput, utilization, and server response times. Based on this feedback, the management server makes intelligent decisions about server provisioning, adding servers only when performance thresholds are exceeded, thereby maintaining productivity while minimizing unnecessary server operations that would increase boot time and migration complexity
Data Source
AI summary
Examples described herein include systems and methods for optimizing the number of servers in a cluster. In one example, a number of application servers, a number of backend servers, and a first disk throughput of a backend server to be included in the cluster are determined. The first disk throughput is determined based on the storage capacity of the backend server and a first round trip time. Example systems and method can also include validating the number of application servers based on a cluster throughput and one of a network interface card bandwidth of an application server to be included in the cluster and a load bearing capacity of the application server. The systems and methods can further include determining a second disk throughput of the backend server and increasing the number of backend servers if the second disk throughput is less than the second disk throughput.