Developing comprehensive performance metrics and optimization strategies for edge computing infrastructures.
Edge faces issues with efficiency, compute availability, and understanding task assigned or requested.
Benchmarks are useful but, existing research is mainly focused on Edge-Cloud pipeline.
Create specific benchmarks based on compute metrics to aid scheduler in Edge Platforms.
Our comprehensive approach to edge computing performance analysis
Collect data on the gpu metrics that are vital in running LLM on Edge Platforms, specfically, metrics such as GPU utilization, VRAM utilization, etc.
Accounts for multiple stages of the compute available, idle stage, overloading prompt requets, and even sending prompt requets while there may background processes running.
Intelligent calculation (weighted sum) using GPU metrics to ouput a single value for testing the capability of the edge platform on using LLMs.