Connor Imes

POET: a portable approach to minimizing energy under soft real-time constraints

Embedded real-time systems must meet timing constraints while minimizing energy consumption. To this end, many energy optimizations are introduced for specific platforms or specific applications. These solutions are not portable, however, and when the application or the platform change, these solutions must be redesigned. Portable techniques are hard to develop due to the varying tradeoffs experienced with different application/platform configurations. This paper addresses the problem of finding and exploiting general tradeoffs, using control theory and mathematical optimization to achieve energy minimization under soft real-time application constraints. The paper presents POET, an open-source C library and runtime system that takes a specification of the platform resources and optimizes the application execution. We test POETs ability to portably deliver predictable timing and energy reduction on two embedded systems with different tradeoff spaces - the first with a mobile Intel Haswell processor, and the second with an ARM big.LITTLE System on Chip. POET achieves the desired latency goals with small error while consuming, on average, only 1.3% more energy than the dynamic optimal oracle on the Haswell and 2.9% more on the ARM. We believe this open-source, library-based approach to resource management will simplify the process of writing portable, energy-efficient code for embedded systems.

Minimizing energy under performance constraints on embedded platforms: resource allocation heuristics for homogeneous and single-ISA heterogeneous multi-cores

This paper explores the problem of energy optimization in embedded platforms. Specifically, it studies resource allocation strategies for meeting performance constraints with minimal energy consumption. We present a comparison of solutions for both homogeneous and single-ISA heterogeneous multi-core embedded systems. We demonstrate that different hardware platforms have fundamentally different performance/energy tradeoff spaces. As a result, minimizing energy on these platforms requires substantially different resource allocation strategies. Our investigations reveal that one class of systems requires a race-to-idle heuristic to achieve optimal energy consumption, while another requires a never-idle heuristic to achieve the same. The differences are dramatic: choosing the wrong strategy can increase energy consumption by over 2× compared to optimal.

Racing and Pacing to Idle: Theoretical and Empirical Analysis of Energy Optimization Heuristics

The problem of minimizing energy for a performance constraint (e.g., Real-time deadline or quality-of-service requirement) has been widely studied, both in theory and in practice. Theoretical models have indicated large potential energy savings, but practical concerns have made these savings hard to realize. Instead, practitioners often rely on heuristic solutions, which achieve good results in practice but tend to be system-specific in efficacy. An example is the race-to-idle heuristic, which makes all resources available until a task completes and then idles. Theory predicts poor energy savings, but practitioners have reported good empirical results. To help bridge the gap between theory and practice, this paper presents a geometrical framework for analyzing the energy optimality of resource allocation under performance constraints. The geometry of the problem allows us to derive an optimal strategy and three commonly used heuristics: 1) race-to-idle, 2) pace-to-idle a near-optimal idling strategy, and 3) no-idle which never idles. We then implement all strategies and test them empirically for seven benchmarks on four different multicore systems, including both x86 and ARM. We find that race-to-idle is near optimal on older systems, but can consume as much as 3× more energy than the optimal strategy. In contrast, pace-to-idle is never more than 12% worse than optimal.

CALOREE: Learning Control for Predictable Latency and Low Energy

Many modern computing systems must provide reliable latency with minimal energy. Two central challenges arise when allocating system resources to meet these conflicting goals: (1) complexity modern hardware exposes diverse resources with complicated interactions and (2) dynamics latency must be maintained despite unpredictable changes in operating environment or input. Machine learning accurately models the latency of complex, interacting resources, but does not address system dynamics; control theory adjusts to dynamic changes, but struggles with complex resource interaction. We therefore propose CALOREE, a resource manager that learns key control parameters to meet latency requirements with minimal energy in complex, dynamic en- vironments. CALOREE breaks resource allocation into two sub-tasks: learning how interacting resources affect speedup, and controlling speedup to meet latency requirements with minimal energy. CALOREE deines a general control system whose parameters are customized by a learning framework while maintaining control-theoretic formal guarantees that the latency goal will be met. We test CALOREEs ability to deliver reliable latency on heterogeneous ARM big.LITTLE architectures in both single and multi-application scenarios. Compared to the best prior learning and control solutions, CALOREE reduces deadline misses by 60% and energy consumption by 13%.

Bard: A unified framework for managing soft timing and power constraints

Embedded systems are subject to timing and power constraints. To support both, software currently must integrate multiple tools, resulting in additional complexity. We address this problem with a unified, portable framework called Bard which uses control theory to meet the primary constraint and linear programming to optimize the other. We evaluate Bard on two embedded platforms that exhibit different performance and power/energy characteristics and show that it achieves less than 2% error in meeting power constraints while maintaining nearly 95% of optimal performance. Additionally, Bard supports changing the primary constraint type at runtime while still achieving similar results.

Portable Multicore Resource Management for Applications with Performance Constraints

Many modern software applications have performance requirements, like mobile and embedded systems that must keep up with sensor data, or web services that must return results to users within an acceptable latency bound. For such applications, the goal is not to run as fast as possible, but to meet their performance requirements with minimal resource usage, the key resource in most systems being energy. Heuristic solutions have been proposed to minimize energy under a performance constraint, but recent studies show that these approaches are not portable - heuristics that are near-optimal on one system can waste integer factors of energy on others. The POET library and runtime system provides a portable method for resource management that achieves near-optimal energy consumption while meeting soft real-time constraints across a range of devices. Although POET was originally designed and tested on embedded and mobile platforms, in this paper we evaluate it on a manycore server-class system. The larger scale of manycore systems adds some overhead to adjusting resource allocations, but POET still meets timing constraints and achieves near-optimal energy consumption. We demonstrate that POET achieves portable energy efficiency on platforms ranging from low-power ARM big.LITTLE architectures to powerful x86 server-class systems.

A portable interface for runtime energy monitoring

As energy consumption becomes a first class concern for computing systems, there is an increasing need for application-level access to runtime power/energy measurements. To support this need, a growing number of power and energy monitors are being developed, each with their own interfaces. In fact, the approaches are extremely diverse, and porting energy-aware code to new platforms with new hardware can involve significant rewriting effort. To reduce this effort and support portable, application-level energy monitoring, a common interface is needed. In this paper, we propose EnergyMon, a portable application interface that is independent of underlying power/energy data sources. We demonstrate EnergyMons flexibility with two case studies -- energy-aware profiling and self-adaptive systems, each of which requires monitoring energy across a range of hardware from different manufacturers. We release the EnergyMon interface, implementations, utilities, and Java and Rust bindings and abstractions as open source.

Energy-efficient Application Resource Scheduling using Machine Learning Classifiers

Resource scheduling in high performance computing (HPC) usually aims to minimize application runtime rather than optimize for energy efficiency. Most existing research on reducing power and energy consumption imposes the constraint that little or no performance loss is allowed, which improves but still does not maximize energy efficiency. By optimizing for energy efficiency instead of application turnaround time, we can reduce the cost of running scientific applications. We propose using machine learning classification, driven by low-level hardware performance counters, to predict the most energy-efficient resource settings to use during application runtime, which unlike static resource scheduling dynamically adapts to changing application behavior. We evaluate our approach on a large shared-memory system using four complex bioinformatic HPC applications, decreasing energy consumption over the naive race scheduler by 20% on average, and by as much as 38%. An average increase in runtime of 31% is dominated by a 39% reduction in power consumption, from which we extrapolate the potential for a 24% increase in throughput for future over-provisioned, power-constrained clusters. This work demonstrates that low-overhead classification is suitable for dynamically optimizing energy efficiency during application runtime.

POET: A Portable Approach to Minimizing Energy Under Soft Real-time Constraints

Embedded real-time systems must meet timing constraints while minimizing energy consumption. To this end, many energy optimizations are introduced for specific platforms or specific applications. These solutions are not portable, however, and when the application or the platform change, these solutions must be redesigned. Portable techniques are hard to develop due to the varying tradeoffs experienced with different application/platform configurations. This paper addresses the problem of finding and exploiting general tradeoffs, using control theory and mathematical optimization to achieve energy minimization under soft real-time application constraints. The paper presents POET, an open-source C library and runtime system that takes a specification of the platform resources and optimizes the application execution. We test POETs ability to portably deliver predictable timing and energy reduction on two embedded systems with different tradeoff spaces - the first with a mobile Intel Haswell processor, and the second with an ARM big.LITTLE System on Chip. POET achieves the desired latency goals with small error while consuming, on average, only 1.3% more energy than the dynamic optimal oracle on the Haswell and 2.9% more on the ARM. We believe this open-source, library-based approach to resource management will simplify the process of writing portable, energy-efficient code for embedded systems.