Dynamic variability management in mobile multicore processors under lifetime constraints

Abstract

Variability is a key issue in modern multiprocessors, resulting in performance and lifetime uncertainty, and high design margins. The margins can be reduced by exposing variability to software and then adapting at runtime. In this work we use sensors to monitor the variable operating conditions and the degradation rate. Based on the sensor data, our variabilityaware OS scheduling algorithm assigns the workload to the cores and sets the power/performance tradeoffs to meet the mobile processor's lifetime constraints while adjusting to variability and improving the overall performance. We implement our algorithm in Android OS on a mobile phone and show that it achieves up to 160% performance improvement over the state-of-the-art while meeting the lifetime constraints.

Performance/Reliability-Aware Resource Management for Many-Cores in Dark Silicon Era

“…Similarly to [8], [18], in some of the experiments, we considered also process variation to analyze its effects on the aging and the capability of the proposed reliabilityaware approach to deal with this additional issue; in particular, we modeled per-core maximum frequency variation map as done also in [18] by means of a normal distribution. For the experiments, we characterized a realistic 12× 12 architecture with a squared floorplan, a chip area of 138mm 2 in 16nm technology, and TDP of 90W.…”

“…At the opposite, as stated in [25], it would be useful to maximize the value of the reliability model within the service period, or at least to fulfill a minimum threshold, to improve the probabilities of the system to not fail before the system is retired. For this reason, other past works [18] adopt an alternative approach for defining the reliability target by setting a given reliability level R(t target ) the system must have at the end of the envisioned lifetime t target . For instance, the reliability target can be specified as follow: At the end of the working life, estimated in t target = 10 years, the system must have at least a reliability of R(t target ) = 45%.…”

“…Many DRM approaches have been proposed for busbased multi-core systems. Some examples are [26], [27], which act on job scheduling to reduce aging effects, and [17], [18], which exploit also DVFS tuning. However, many-core systems have a very different architecture, programming paradigm and execution model from the shared-memory bus-based multi-core counterpart.…”

CHIPS-AHOy

“…We then customize and incorporate the technique presented in [18] to provide the scheduler with reliability awareness. To indicate the expected lifetime and to measure system reliability during its operational life, we express a minimum reliability level R t ar дet the system must fulfill at the end of the service life t t ar дet (as in [29]).…”

VarDroid

“…Recent publications on Dynamic Variability Management (DVM) propose scheduling and DVFS algorithms for embedded platforms [21] and for Android-based devices [19]. However, these solutions do not account for real workload and real user interaction.…”