← Back to the chapter · Previous detail · Next detail →

Technical detail · Compute Offloading and Placement

LEO-style heterogeneous scheduling

15.11 LEO-Style Heterogeneous Scheduling

Follow the sensor application into the sensor-job buffer, where jobs wait in deadline order. The workload monitor tracks queued jobs and recent runtimes. The resource monitor supplies processor, memory, thermal and radio state. Both feed the low-power scheduler, which scores feasible targets for energy, lateness and service risk. Its dispatch arrow leads to CPU, DSP, GPU or an off-device target. A dashed red path marks fallback after a timeout or unavailable accelerator. Completion measurements loop back to update the models. The bottom strip reports a historical 10-app workload with about 100 ms scheduling time and less than 0.5% workload energy.

  1. Sensor applications describe jobs with data size, deadline, accuracy mode, and permitted execution targets.
  2. A workload monitor observes the queued jobs and their recent runtimes.
  3. A resource monitor records processor availability, frequency, memory pressure, thermal state, radio state, and the measured energy model for CPU, DSP, GPU, or cloud execution.
  4. Jobs wait in a sensor-job buffer so compatible work can be batched without losing deadline order.
  5. A scheduler running on a low-power unit scores feasible placements and dispatches each job to a CPU, DSP, GPU, or off-device target.
  6. Completion measurements update the models; a timeout or unavailable accelerator takes a declared fallback path.

For job jj on processing element kk, a useful score is not runtime alone:

Sj,k=Ej,k+λLmax⁡(0,tj,k−dj)+λQRj,k,S_{j,k}=E_{j,k}+\lambda_L\max(0,t_{j,k}-d_j)+\lambda_Q R_{j,k},

where Ej,kE_{j,k} includes migration and data movement, djd_j is the deadline, and Rj,kR_{j,k} represents a quantified service risk such as unsupported precision or cloud unavailability. Infeasible placements receive infinite score. The scheduler then chooses the lowest-score feasible target, rechecking shared memory and thermal constraints before dispatch.

Historical keyword-spotting experiments illustrate why the control path must stay cheap. In LEO’s tested sensor-app workload, the low-power scheduler took about 100 ms on average and used less than 0.5% of the phone’s daily battery. LEO study These figures are workload- and platform-specific; the durable requirement is Eschedule≪EworkE_{schedule}\ll E_{work} and tschedulet_{schedule} small relative to the tightest deadline.

← Back to the chapter · Previous detail · Next detail →