Flexibility in the Cloud Elevates Innovation and Research Impact
The Wharton School uses WEKA® NeuralMesh™ to adjust its HPC3 cluster resources flexibly based on demand.
Introduction
Faster Time to Insights Drives Ground-Breaking Research
The Wharton School of the University of Pennsylvania fosters a legacy of ground-breaking innovation and collaboration by supporting its award-winning faculty and graduate students as they pursue cutting-edge research in fields such as AI and Analytics, Behavioral Economics, Finance, Leadership, and Public Policy.
The Wharton experience is defined by their strategic vision of data-driven approaches and scalable solutions to elevate their research impact throughout the world. With over 20 research centers, 225 faculty members, and more than 150 graduate students, there is significant pressure to utilize a data infrastructure that is highly performant and flexible enough to support a breadth of research and intensive-compute workloads.
Architecting and managing a data infrastructure that can meet the ongoing research needs of such a large, diverse community, however, is not without its challenges. In the increasingly fast-paced world of academic publication, reducing the time that it takes to run workloads can lead to measurably shorter time to insights and the ability to not only meet deadlines, but exceed them and be the first to publish ground-breaking research in a field.
Going from the old NFS solution to NeuralMesh has been such an upgrade. The amount of throughput we get allows researchers to meet their publication deadlines.
The Challenge
Struggling with Efficiency and Scalability
Research projects at The Wharton School have grown in size and complexity, which led to challenges for the existing data infrastructure that struggled with efficiency and scalability. As data management became more challenging, they began to search for a solution that would streamline their data processes in the cloud.
- Complicated cloud data management
- Performance at scale
- Inflexible resource allocation
HPC tends to turn CPU problems into I/O problems. NeuralMesh turned those I/O-bound problems back into CPU-bound problems.
The Solution
The Scale and Economics of Cloud Object Stores with the Speed of SSD
NeuralMesh™ combines a wide range of storage mechanisms and the cloud that provides an extendable file system offering the scale and economics of cloud object stores and the speed of SSD and dedicated networking.
NeuralMesh™ combines a wide range of storage mechanisms and the cloud to provide an extendable file system offering the scale and economics of cloud object stores and the speed of SSD and dedicated networking.
With NeuralMesh, we can expand input/output capabilities to ensure that even the most data-intensive research projects are supported efficiently.
The Outcome
Realizing the Promise of Elastic Cloud Computing
Wharton leverages NeuralMesh’s elasticity and easy configuration and management to their advantage. Reducing the time that it takes to run workloads ensures that researchers meet publication deadlines.
- Powerful Automation: Automation and policy-driven management simplifies administration of the HPC3 cluster.
- Increased Throughput: Increased throughput to every compute node allows significantly better time to solution.
- Increased Productivity: Realizing the promise of elastic cloud computing allows us to grow and run as many jobs as our researchers require.
Scale Production AI Faster with NeuralMesh
Your models aren't slow. Your data is. Fix AI bottlenecks with high-throughput infrastructure.