LaLaRAND accepted at IEEE RTSS 2021

Our paper “LaLaRAND: Flexible Layer-by-Layer CPU/GPU Scheduling for Real-Time DNN Tasks” was accepted at the 42nd IEEE Real-Time Systems Symposium.

LaLaRAND architecture showing layer-level CPU and GPU execution coordinated by a system-wide runtime scheduler — LaLaRAND coordinates layer-level CPU/GPU execution through a system-wide scheduler.

LaLaRAND coordinates DNN execution at layer granularity across heterogeneous CPU and GPU resources. It combines quantization-aware execution with system-wide scheduling to improve schedulability while limiting accuracy loss.

The paper is available from IEEE Xplore.

Source: RTCL@DGIST announcement