NexusIgnite Storage Services

S3 Service

Simple Storage Service (S3) is a scalable, reliable, and cost-effective object storage service designed for storing and retrieving large amounts of data. It offers durability by redundantly storing data across multiple devices, ensuring high availability and robust security features, including encryption and fine-grained access control. S3 has one high-performance tier that allows fast data access and retrieval at an affordable cost, all while seamlessly integrating with other services.

NFS Service

Network File System (NFS) is a fully managed, scalable, and elastic file storage service designed to be mounted on Processing Units simultaneously. NFS offers a simple and highly available file system for Linux-based workloads, automatically scaling up or down as files are added or removed, ensuring efficient use of storage capacity. It provides robust security features, including encryption at rest and in transit, and integrates seamlessly with other services. NFS is ideal for various use cases such as web serving, content management, data analytics, and container storage, offering a cost-effective solution with high performance.

SSMP Service

Single Storage Multi-Protocol (SSMP) is a fully managed, scalable storage service accessible through S3 or NFS and designed to be mounted on multiple GPU instances simultaneously. SSMP users can write data using S3 APIs and read it using the NFS protocol, or vice versa. SSMP offers a simple and highly available file system for Linux-based workloads directly connected to S3 (with NFS performance), automatically scaling up or down as files are added or removed, ensuring efficient use of storage capacity. It provides robust security features, including encryption at rest and in transit, and integrates seamlessly with other services. SSMP is ideal for use cases such as data analytics, machine learning, and container storage.

RDMA-Enabled Storage Service

RDMA-enabled storage is a fully managed, scalable, multi-protocol storage service designed for direct GPU access using RDMA. RDMA storage offers a simple and highly available file system for Linux-based workloads directly connected to S3 (with NFS performance), automatically scaling up or down as files are added or removed, ensuring efficient use of storage capacity. It provides robust security features, including encryption at rest, and integrates seamlessly with other services. RDMA storage is ideal for GPU-accelerated AI workloads.

Feature Store Service

Feature Store is a fully managed repository for storing, retrieving, and managing machine learning and AI features. It simplifies the process of creating, updating, and sharing features for machine learning models, ensuring consistency and accuracy across training and inference. Feature Store gives GPUs direct access to the feature store using RDMA, making it suitable for large-scale data processing and model training. Integrated with other services, it offers robust security, scalability, and seamless data management, making it easier to build, deploy, and manage machine learning workflows effectively.