Batch Job QoS Control via Dynamic Resource Reservations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face challenges in efficiently managing resource allocations for batch jobs to meet quality of service (QoS) specifications, particularly in ensuring timely completion while optimizing resource usage and minimizing resource wastage.
Innovation Solution
The implementation of a batch job QoS control algorithm that dynamically adjusts resource reservations over time slices based on tolerated failure probabilities, separating batch job QoS concerns from data center resource optimization components, and using statistical packing algorithms to optimize resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data centers allocate sufficient resources to meet QoS specifications for batch jobs, then job completion reliability is improved, but resource usage efficiency deteriorates due to over-provisioning
Solution Approach 1:
The patent implements dynamic resource allocation where the resource reservation level is adjusted over time based on the batch job's progress and remaining deadline. The system transitions from conservative initial reservations to more aggressive allocations as the job progresses, allowing the allocation strategy to adapt to actual job behavior and remaining time, thus improving both reliability and efficiency
Solution Approach 2:
The system changes the resource allocation parameter dynamically based on the time slice and job progress. By modifying the reservation level parameter according to the batch job's execution state and tolerated failure probability, the system optimizes the balance between meeting QoS requirements and efficient resource utilization
2Device complexity
If data centers use static resource allocation for batch jobs, then implementation simplicity is improved, but QoS specification adherence deteriorates due to inability to adapt to varying job requirements
Solution Approach 1:
The patent segments the batch job execution into discrete time slices and applies different resource allocation strategies to each segment. By dividing the execution timeline into manageable intervals with specific reservation levels, the system achieves adaptive QoS management without requiring complex continuous control mechanisms
Solution Approach 2:
The system incorporates feedback mechanisms where the resource allocation for subsequent time slices is determined based on the progress and performance observed in previous slices. This feedback loop allows the system to adapt to actual job behavior patterns while maintaining relatively simple allocation rules
3Loss of time
If data centers increase resource reservations to ensure timely batch job completion, then job completion timeliness is improved, but resource wastage increases due to reserved resources remaining unused
Solution Approach 1:
The patent applies periodic resource allocation adjustments at each time slice boundary. By reviewing job progress periodically and resetting the reservation level for the next period based on current state and remaining deadline, the system ensures timely completion while minimizing persistent over-provisioning that would cause resource wastage
Data Source
AI summary
A data center management system can include a data center customer profile corresponding to a data center customer, where the data center customer profile includes a service level agreement (SLA) and a data center resource usage model. A batch job quality of service (QoS) control module can determine an immediate QoS specification for a batch job of the data center customer, and a data center resource optimization module can determine a data center resource allocation for the data center customer based on the data center customer profile and the immediate QoS specification.