Ricardo Macedo

José Peixoto, Alexis Gonzalez, Janki Bhimani et al.

Programmable caching engines like CacheLib are widely used in production systems to support diverse workloads in multi-tenant environments. CacheLib's design focuses on performance, portability, and configurability, allowing applications to inherit caching improvements with minimal implementation effort. However, its behavior under dynamic and evolving workloads remains largely unexplored. This paper presents an empirical study of CacheLib with multi-tenant settings under dynamic and volatile environments. Our evaluation across multiple CacheLib configurations reveals several limitations that hinder its effectiveness under such environments, including rigid configurations, limited runtime adaptability, lack of quality-of-service support and coordination, which lead to suboptimal performance, inefficient memory usage, and tenant starvation. Based on these findings, we outline future research directions to improve the adaptability, fairness, and programmability of future caching engines.

Rahma Nouaji, Stella Bitchebe, Ricardo Macedo et al.

Data loaders are used by Machine Learning (ML) frameworks like PyTorch and TensorFlow to apply transformations to data before feeding it into the accelerator. This operation is called data preprocessing. Data preprocessing plays an important role in the ML training workflow because if it is inefficiently pipelined with the training, it can yield high GPU idleness, resulting in important training delays. Unfortunately, existing data loaders turn out to waste GPU resources, with $76\%$ GPU idleness when using the PyTorch data loader, for example. One key source of inefficiency is the variability in preprocessing time across samples within the same dataset. Existing data loaders are oblivious to this variability, and they construct batches without any consideration of slow or fast samples. In this case, the entire batch is delayed by a single slow sample, stalling the training pipeline and resulting in head-of-line blocking. To address these inefficiencies, we present MinatoLoader, a general-purpose data loader for PyTorch that accelerates training and improves GPU utilization. MinatoLoader is designed for a single-server setup, containing multiple GPUs. It continuously prepares data in the background and actively constructs batches by prioritizing fast-to-preprocess samples, while slower samples are processed in parallel. We evaluate MinatoLoader on servers with V100 and A100 GPUs. On a machine with four A100 GPUs, MinatoLoader improves the training time of a wide range of workloads by up to $7.5\times$ ($3.6\times$ on average) over PyTorch DataLoader and Pecan, and up to $3\times$ ($2.2\times$ on average) over DALI. It also increases average GPU utilization from 46.4\% with PyTorch to 90.45\%, while preserving model accuracy and enabling faster convergence.