Data Storage for Neural Network Training

A global leader in automotive design and production required a solution to store sensor data that is used to train safety-critical neural network algorithms. The company required a low total cost of ownership (TCO) so they could manage the solution without adding support resources. With limited space in their data center they also needed a solution with a small footprint and no additional HVAC requirements.

Although they already had a Hadoop cluster in place the company chose to invest in a Quantum storage solution via AutonomoStuff underscoring the value they saw in Quantum.

Challenge:

• Quick access to recently collected data from vehicles
• Large datasets stored for long time periods at a low cost per TB
• Limited space available in the data center
• Minimal heat generation to allow management with existing data center HVAC
• Support for a variety of communication protocols including Ethernet iSCSI and fiber channel
• High-performance support for engineering testing and analytics computing
• Constrained resources requiring a low TCO solution—both operational and environmental

Solution:

• Quantum XcellisTM Workflow Director powered by StorNext
• 900 TB: StorNext QXSTM-5600 performance storage
• 2.4 PB: StorNext AELI6 archive storage 200 slots
• Total Storage: 3.3PB

Customer benefits:

• Simplified workflow with shared storage eliminating storage silos
• Faster performance support for high-performance computing (HPC) analytics and accelerated development
• 29u rack space for the entire storage solution
• No additional heat generated in the data center thanks to tape storage
• Support for multiple communication protocols for varied users and devices
• Scalable performance and capacity to address future needs and control budget
• Options to non-disruptively expand to the cloud and other storage technologies in the future
• No additional on-site resourcing required for deployment and management
• Total cost: 1/10 the cost of the alternative solution