What if your cloud analytics environment was smart enough to spin up the exact amount of computing resources required for any specific job?
What if you had an AI tool that could determine your computing needs for any given analytics project, scale up to run the job and then scale down when the project was done running?
Traditional computing environments are prone to over-allocating resources to handle worst-case demand scenarios, leading to underutilization during normal or low-demand periods. Then, when high demand is needed, users submit requests to IT to ask for more resources to process and complete their work.
One of the unique challenges of managing analytic workloads is that resource needs for a decision tree, for example, differ from resource needs for a random forest. Resource needs depend on the technique being used, plus the depth and breadth of data, memory utilization, CPU or GPU usage, how the models are threaded – and more. Resourcing decisions involve multiple inputs and calculations that make it exceedingly complex for a human to reconcile accurately and translate into an optimal deployment.
Revolutionize analytic cloud workloads with AI-driven resource management
A new patent from SAS proposes a system where computing resources are dynamically allocated based on real-time demand, informed by a machine learning feedback sequence. This approach aims to optimize resource use and reduce waste.
Rich Wellum is a manager of the Cloud Analytics Server at SAS and one of the inventors. He says the patented technique uses AI to dynamically decide deployment configurations based on analytical needs, enhancing cloud resource management without manual IT intervention. The system would rely on historical data and AI to optimize deployment configurations for all types of jobs. John Hardin Gelpi and Alexander Daehnrich are co-inventors.
“The AI-driven deployment configuration could simplify scaling and resource management and reduce the need for manual IT intervention,” explains Wellum.
To illustrate one potential use of the patented technique, Wellum describes a large bank that performs analytics with two terabytes of data every Tuesday at 4:00 a.m. Currently, an administrator has to manually spin up the necessary infrastructure the night before and monitor the job. With the patented AI system, the infrastructure would be automatically prepared at 1:00 a.m. on Tuesday, based on the job history, eliminating the need for manual intervention.
The patent, titled "Systems and Methods for Dynamic Allocation of Compute Resources via a Machine Learning-Informed Feedback Sequence," offers several benefits, including:
- Optimized resource utilization: By dynamically allocating computing resources based on specific project needs, the system ensures that resources are used efficiently. This reduces the likelihood of over-provisioning and underutilization, leading to cost savings and better performance.
- Reduced operational costs: Since the system can adjust resource allocation in real time, it minimizes the need for excess capacity. This can lead to substantial cost reductions in hardware, energy consumption and maintenance.
- Improved performance: The machine learning-informed feedback sequence allows the system to predict and respond to changes in demand more accurately. This results in improved performance and responsiveness, especially during peak usage times.
- Scalability: The dynamic nature of the system makes it highly scalable. It can easily adapt to varying workloads and can be implemented in different environments, from small-scale applications to large enterprise systems.
- Enhanced user experience: By ensuring that compute resources are available when needed, the system can provide a smoother and more reliable user experience. This is particularly beneficial for applications that require high availability and quick response times.
These benefits make the patented system a valuable innovation for organizations looking to optimize their compute resource management and improve overall efficiency.
More machine learning patents from SAS
SAS inventors have received patents for many other AI innovations. Recent patents cover a range of techniques and applications that primarily focus on advanced data processing, machine learning, and natural language processing technologies.
Other patents include a new technique for identifying near-duplicate text documents and a new method for combining large language models with natural language processing to improve generative AI output.
Collectively, these SAS patents illustrate a dedication to pushing the boundaries of machine learning and using AI in new ways to improve decision making and efficiency.
