AI Integration in Business: A Case Study of NetApp ONTAP AI in AI_LAB

• Infrastructure as a Service (IaaS) - Hybrid Cloud
• Infrastructure as a Service (IaaS) - Public Cloud

• Buildings
• Education

• Human Resources
• Quality Assurance

• Clinical Image Analysis
• Intelligent Packaging

• Cloud Planning, Design & Implementation Services
• Training

ITOCHU Techno-Solutions (CTC) operates the Technical Solution Center (TSC), a comprehensive facility that allows client companies to evaluate and verify systems to be introduced using actual equipment. It is Japan's largest verification facility for a multi-vendor environment and offers full technical support from experienced engineers. In recent years, the TSC has seen a significant increase in the number of AI/deep learning verification projects, and opportunities to verify data and service integrations between the TSC system and public cloud services like AWS, Microsoft Azure, and Google Cloud Platform. This growing need for verification of hybrid cloud and AI/deep learning drove CTC to create a new verification environment, AI_LAB, in late 2018. AI_LAB was designed to verify large-scale AI utilization and to build learning models. However, the challenge was to establish a system that could handle the advanced verification requirements of AI/deep learning.

ITOCHU Techno-Solutions

The customer in this case study is ITOCHU Techno-Solutions (CTC), a company that operates the Technical Solution Center (TSC), a comprehensive facility that allows client companies to evaluate and verify systems to be introduced using actual equipment. CTC is dedicated to providing a multi-vendor comprehensive verification facility that offers full technical support from experienced and dedicated engineers. The company conducts approximately 1,200 performance evaluations and functional tests annually using the latest system products. In recent years, CTC has seen a significant increase in the number of AI/deep learning verification projects and opportunities to verify data and service integrations between the TSC system and public cloud services.

To meet the challenge, CTC adopted the NetApp ONTAP AI integrated system, jointly developed by NetApp and NVIDIA, for the AI-dedicated verification environment AI_LAB. The system integrates the NVIDIA DGX-1 AI, incorporating eight Tesla V100s and the NetApp AFF A800 high-end all-flash array, forming a reference architecture optimized for AI/deep learning applications. The configuration adopted for CTC's AI_LAB includes the NVIDIA DGX-1 cluster with 4 nodes/32 GPUs and one NetApp AFF A800 system provided with a high-speed connection by two 100 GbE switches. The NetApp AFF A800 equipped with NVMe SSD can supply training data to the GPUs at an extremely high rate, making it possible to take full advantage of the petaflops computational performance of the DGX-1 cluster. AI_LAB also integrates with public cloud services such as AWS, Azure, and GCP, allowing customers to easily build an AI/deep learning data pipeline by connecting the on-premises verification environment in AI_LAB to the public cloud. Furthermore, AI_LAB provides an environment with the capability to meet customers’ every need, from large-scale verification that makes full use of 32 GPUs to small-scale tests with 1 GPU.

• The adoption of NetApp ONTAP AI in AI_LAB has significantly contributed to the utilization of AI for business by client companies. The system's integration with public cloud services allows customers to easily build an AI/deep learning data pipeline, enabling not just verification of cooperative processing using GPUs on the cloud side and hierarchical management of data that incorporates cloud storage, but also verification of the return on investment of on-premises and cloud AI environments. Furthermore, the system's high level of compatibility with container environments, such as the provision of NetApp Trident, a tool that makes it easy to create the “persistent volumes” that are essential for handling applications and databases in a container environment, allows for the movement of applications that incorporate the learning model into the cloud and run them there. This has resulted in an environment in which the customer can use the necessary data and computational resources immediately on demand, maximizing the time customers spend on verification and increasing the efficiency of AI_LAB resource utilization.

• Throughput of up to 25 GB/s achieved in performance verification of NetApp ONTAP AI
• Latency of 500 microseconds or less while constantly maintaining utilization rate of at least 95% on all 32 GPUs of the DGX-1 cluster
• AI_LAB is always fully booked at least 3 months in advance, indicating high demand and utilization