Power Management Algorithm for Throughput Workloads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing power management algorithms for System-on-Chip (SoC) fail to differentiate between throughput and latency critical workloads, leading to sub-optimal power efficiency and performance, as they either focus on individual processors or use ad-hoc solutions that do not account for global interactions between SoC and workloads.
Innovation Solution
A power management algorithm framework that determines processing core activity deviation data to identify homogeneous workloads, adjusts core frequencies based on binomial distribution analysis, and uses a Quality-of-Service (QoS) metric to differentiate between workload types, optimizing power efficiency without performance degradation by reducing idle time in throughput-based workloads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If existing power management algorithms are used that focus on individual processors or use ad-hoc solutions, then device complexity is reduced, but power efficiency deteriorates because they cannot differentiate between throughput and latency critical workloads
Solution Approach 1:
The patent segments the workload analysis by differentiating between throughput-based and latency-critical workloads using distinct metrics (throughput metric vs. latency metric). This segmentation allows the power management algorithm to apply different frequency adjustment strategies to different workload types, improving power efficiency without requiring a single complex universal algorithm.
Solution Approach 2:
The patent introduces an intermediary mechanism - a global QoS objective function that mediates between individual processor states and overall system performance. This intermediary enables coordinated power management across multiple processors while maintaining system-wide QoS, resolving the contradiction between simple individual processor management and complex system-wide optimization.
2Use of energy by moving object
If frequency is lowered to minimize idle times in throughput-based workloads, then power efficiency improves, but performance may degrade if not properly differentiated from latency critical workloads
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously monitors workload characteristics (throughput metric vs. latency metric) and adjusts processor frequency accordingly. For throughput-based workloads, frequency is lowered when idle time is detected, while for latency-critical workloads, frequency is maintained. This feedback loop ensures power efficiency improvements without performance degradation.
Solution Approach 2:
The patent changes the operational parameters of processors dynamically based on workload type identification. Throughput-based workloads receive frequency reduction parameters when idle, while latency-critical workloads maintain high frequency parameters. This parameter adaptation resolves the contradiction by making frequency adjustment conditional on workload characteristics rather than applying a single parameter set.
3Productivity
If local optimization algorithms are applied to individual processors, then device complexity remains low, but productivity deteriorates due to poor global QoS optimization
Solution Approach 1:
The patent merges individual processor optimization with global QoS optimization by combining local throughput/latency metric collection with a global objective function. Rather than separate local and global algorithms, the system integrates them into a unified framework where local metrics feed into global decisions, improving productivity while keeping coordination complexity manageable through modular integration.
Solution Approach 2:
The patent creates a universal power management framework that handles both throughput-based and latency-critical workloads through a single multi-functional algorithm. This universal algorithm performs workload classification, frequency adjustment, and QoS monitoring in one integrated system, improving global productivity without requiring multiple separate optimization mechanisms.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A power management algorithm framework proposes: 1) a Quality-of-Service (QoS) metric for throughput-based workloads; 2) heuristics to differentiate between throughput and latency sensitive workloads; and 3) an algorithm that combines the heuristic and QoS metric to determine target frequency for minimizing idle time and improving power efficiency without any performance degradation. A management algorithm framework enables optimizing power efficiency in server-class throughput-based workloads while still providing desired performance for latency sensitive workloads. The power savings are achieved by identifying workloads in which one or more cores can be run at a lower frequency (and consequently lower power) without a significant negative performance impact.