Light thread placement cuts energy on multi core processors
Ride, Raise, or Wake: A Theory of Light-Thread Placement on Multi-Tier Processors
Performance
Summary
Modern phones and laptops have different types of processor cores, each with its own power settings. When a small task needs to run, it’s tricky to decide whether it should share a fast, already-awake core or wake up a more energy-efficient core. The authors introduce a clear rule to choose which core to use based on the task's demand and power costs, balancing energy use. They measure this on a real Apple chip to confirm their model.
What this means in practice
- •For chip architects: Optimize energy use in multi-core processors by applying a simple cost-based rule to place light threads efficiently across core tiers.
- •For operating system developers: Improve scheduler decisions by integrating tier wake costs and energy-per-operation metrics to reduce power use when assigning threads to cores.
Authors
Mohamed Amine Bergach
Abstract
hones and laptops now mix several kinds of cores, and each kind sits in its own voltage and clock domain. When a light thread arrives, should it run beside heavy threads on a fast cluster that is already awake, or wake a cluster of efficient cores? We answer with three costs. A thread can \emph{ride} an awake domain at the voltage other threads hold, \emph{raise} that voltage for everyone, or \emph{wake} a sleeping domain and pay its fixed power. From these costs follows a simple rule: ride while the thread's demand is below $r^*=Φ/(e_h-e_l)$, the fixed power of the cluster it would wake over the gap in energy per operation; otherwise wake the smallest tier that can serve it. The best tier is not monotone in demand: tiny threads ride high, medium threads wake a low tier, heavy threads go high again. Two common designs err on opposite sides of the rule: grouping a light thread with the heavy threads of its application overpays above the threshold, and an energy model without idle costs overpays below it. In a model with five tiers, grouping costs 80 to 145\% more energy than the rule, and more as tiers are added. Measurements on an Apple M6 anchor the parameters.