Intelligent Management of Mobile Systems Through Computational Self-Awareness

Introduction

Battery powered-devices are the most ubiquitous computers in the world. Users expect the devices to support high performance applications running on same device, sometimes at the same time. The devices support a wide range of applications, from interactive maps and navigation, to web browsers and email clients. In order to meet the performance demands of the complex workloads, increasingly powerful hardware platforms are being deployed in battery-powered devices. These platforms include a number of configurable knobs that allow for a tradeoff between power and performance, e.g., dynamic voltage and frequency scaling (DVFS), core gating, idle cycle injection, etc. These knobs can be set and modified at runtime based on workload demands and system constraints. Heterogeneous manycore processors (HMPs) have extended this principle of dynamic power-performance tradeoffs by incorporating single-ISA, architecturally differentiated cores on a single processor, with each of the cores containing a number of independent tradeoff knobs. All of these configurable knobs allow for a large range of potential tradeoffs. However, with such a large number of possible configurations, HMPs require intelligent runtime management in order to achieve application goals for complex workloads while considering system constraints. Additionally, the knobs may be interdependent, so the decisions must be coordinated. In this chapter, we explore the use of computational self-awareness to address challenges of adaptive resource management in mobile multiprocessors.

Computational Self-awareness

Self-aware computing is a new paradigm that does not strictly introduce new research concepts, but unifies overlapping research efforts in disparate disciplines (Lewis et al., 2016). The concept of self-awareness from psychology has inspired research in autonomous systems and neuroscience, and existing research in fields such as adaptive control theory support properties of self-awareness. This chapter addresses key challenges for achieving computational self-awareness that can make the design, maintenance and operation of complex, heterogeneous systems adaptive, autonomous, and highly efficient. Computational self-awareness is the ability of a computing system to recognize its own state, possible actions and the result of these actions on itself, its operational goals, and its environment, thereby empowering the system to become autonomous (Jantsch et al., 2017). An infrastructure for system introspection and reflective behavior forms the foundation of self-aware systems.

Reflection

Reflection can be defined as the capability of a system to reason about itself and act upon this information (Smith, 1982). A reflective system can achieve this by maintaining a representation of itself (i.e., a self-model) within the underlying system, which is used for reasoning. Reflection is a key property of self-awareness. Reflection enables decisions to be made based on both past observations, as well as predictions made from past observations. Reflection and prediction involve two types of models: (1) a self-model of the subsystem(s) under control, and (2) models of other policies that may impact the decision-making process. Predictions consider future actions, or events that may occur before the next decision, enabling “what-if” exploration of alternatives. Such actions may be triggered by other resource managers running with a shorter period than the decision loop. The top half of Figure 1 shows prediction enabled through reflection that can be utilized in the decision making process of a feedback loop. The main goal of the prediction model is to estimate system behavior based on potential actuation decisions. This type of prediction is most often performed using linear regression-based models (Mück et al., 2015; Pricopi et al., 2013; Annamalai et al., 2013; Singh et al., 2009) due to their simplicity, while others employ a binning-based approach in which metrics sensed at runtime are used to classify workloads into categories (Liu et al., 2013; Donyanavard et al., 2016).

Figure 1.

Feedback loop overview. The bottom part of the figure represents a simple observe-decide-act loop. The top part adds the reflection mechanism to this loop, enabling predictions for smart decision making.