Скачать или смотреть High-Performance Storage for Machine Learning

Use cases like machine learning, deep learning and artificial intelligence all require massive amounts of performance. These systems use dedicated GPU accelerators for training machine learning algorithms and massive amounts of network bandwidth. These environments also push today’s all-flash systems to their breaking point. NVMe flash systems will help, but the problem is that most systems on the market today are block-based where most of these deep learning applications are designed for file systems. The problem here is most file systems are decades old and not optimized to take advantage of these new architectures. They are also ignorant of cloud resources.

WekaIO CEO/Co-founder Liran Zvibel joins Storage Switzerland for a ChalkTalk discussion on the limitations of current file systems and the need for a new file system that is optimized for NVMe and can leverage the cloud to burst workloads there.