Hybrid-Smash: a heterogeneous CPU-GPU compression library
[EN] Compression algorithms are widely used to reduce data size and improve application performance. Nevertheless, data compression has a computational cost which can limit its use. GPUs could be leveraged to reduce compression time. However, existing GPU-based compression libraries expect data to c...
| Autores: | , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2024 |
| País: | España |
| Institución: | Universitat Politècnica de València (UPV) |
| Repositorio: | RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia |
| Idioma: | inglés |
| OAI Identifier: | oai:riunet.upv.es:10251/208656 |
| Acceso en línea: | https://riunet.upv.es/handle/10251/208656 |
| Access Level: | acceso abierto |
| Palabra clave: | Lossless compression Parallel computing GPU ARQUITECTURA Y TECNOLOGIA DE COMPUTADORES |
| Sumario: | [EN] Compression algorithms are widely used to reduce data size and improve application performance. Nevertheless, data compression has a computational cost which can limit its use. GPUs could be leveraged to reduce compression time. However, existing GPU-based compression libraries expect data to compress in GPU memory, although it is usually stored in CPU memory. Additionally, setup time of GPUs could be a problem when compressing small data sizes. In this paper, we implement a newGPU-based compression library. Contrary to existing ones, our library uses data located in CPU memory. Performance results show that, for the same compression algorithms, GPUs are beneficial for larger data sizes whereas smaller data sizes are compressed faster using CPUs. Therefore, we enhance our proposal with Hybrid-Smash: a heterogeneous CPU-GPU compression library, which transparently uses CPU or GPU compression depending on data size, thus improving compression for any data size. |
|---|