How IoT Analytics Can Benefit from GPUs and GPU Databases
If your organization has jumped on the IoT bandwagon, it’s most likely already dealing with structured and unstructured data from either consumer devices and sensors or from industrial IoT data.
Correlating data feeds from multiple sources is a tough task for any organization. Some companies have turned to Hadoop and other open source software, but these are often limited by scale, I/O or compute.
GPU computing offloads compute-intensive portions of the application to the GPU, while the remainder of the code runs on the CPU. As a result, users experience a dramatic throughput boost for their IoT workloads.
Existing Solutions Aren’t Working
Your company may have moved from legacy to newer commercial or open source technologies to take advantage of IoT data. This typically involves “duct taping” 5-10 open source and commercial technologies together for a workable solution. These moves are often quite costly and complex, take months to move to production and rely on batch processing instead of real-time processing.
GPU Databases Are the Answer for IoT Analytics
GPU databases bring revolutionary capabilities to IoT data and analytics. First, NVIDIA GPUs take traditional database operations and accelerate them by using thousands of small, efficient cores that are well-suited to performing repeated similar instructions in parallel.
Some of the latest GPUs feature over 4,000 cores versus just 16 to 32 cores on a typical CPU-based device. Additionally, GPU cores can crunch data far more efficiently and faster than CPUs, which process data sequentially.
These features make GPUs ideal for analyzing massive datasets in real time, particularly for use cases where time and location matter.
Look for a GPU database that has native GIS and IP-address object types, geospatial functions and map-based visualizations built into it. These enable GPUs to render on-the-fly pictures and videos that are created at the time of query.
You’ll also want extensibility with user-defined functions to integrate custom code and open source machine learning libraries such as TensorFlow, Caffe and Torch to train and run various machine learning and deep algorithms.
The GPU database should be able to linearly scale out on premise or in the cloud. It’s also important that the GPU database has connectors for Hadoop, Spark, Kafka, NiFi, Accumulo and Kibana, so you can read/write data-in-motion and data-at-rest, and the GPU database can function as the fast layer on top of your data stores.
Additionally, connectors and integration with business intelligence tools, like Tableau or Power BI, can make the information easily and more quickly accessible to the business.
IoT Use Cases
There have been dozens of consumer and industrial IoT use cases deployed, with more to come. Some of the top use cases include:
Fleet Management – The proliferation of mobile networks has enabled rapid growth in fleet management systems. A prime example of how GPU databases can be used as a real-time analytics platform for fleet management is the United States Postal Service, which deploys Kinetica to optimize its business operations.
USPS mail carriers use a device that scans packages and emits their exact geographic location every minute. With Kinetica, USPS can collect, process and analyze over 200,000 messages per minute. It can analyze this breadcrumb data to understand where spending achieves the best results, make faster and more efficient strategic decisions, provide customers with more reliable service and reduce costs by streamlining deliveries.
Since Kinetica’s GPU database can be used to visualize geospatial data, dispatchers can efficiently plan territory assignments and make better use of routes.
Smart Grid – Energy and public utility companies are using “smart meters” to measure how energy resources are being used in homes and commercial buildings. These systems help utilities meet the demand for energy conservation, while also making billing easier for customers to understand.
Homeowners who use smart meters have real-time visibility into their energy consumption and can adjust accordingly, while utilities are better able to meet consumer demand and balance production. In addition, smart meters can continuously monitor energy use, so utilities can react quickly to broken equipment or service interruptions.
Manufacturing – A key challenge facing manufacturing IoT is keeping up with the ingestion of streaming events coming from assembly lines, the supply chain, the manufacturing machines themselves, as well as the tags on individual items being produced.
A GPU database provides the necessary performance requirements for today’s many manufacturing IoT use cases, including optimizing the entire manufacturing chain, performing streaming analytics on component functionality, tracking and monitoring inventory, materials and operations, detecting manufacturing defects, ensuring safety and avoiding failures, and tracking quality, returns and warranty claims.
Customer Experience – GPU databases can be deployed at retailers to track and analyze huge volumes of moving assets and inventory in real time – ideal for generating faster and more relevant intelligence across a company’s supply chain.
The classic IoT use case for retailers is customer 360: retailers can correlate data from point-of-sale systems, social media streams, weather forecasts and wearable devices. This data is used to develop a better understanding of customers and the business by being able to query massive datasets in seconds vs. hours.
Supply Chain Optimization – GPU databases can be used to provide real-time, location-based insights across the entire supply chain, including suppliers, distributors, logistics, transportation and retail locations for businesses to understand demand, manage supply and track inventory in real time.
Delivering Speed, Scale and Intelligence to IoT
NVIDIA GPUs and GPU databases deliver speed, scale and intelligence for IoT by converging machine learning and online analytical processing. This enables real-time insights to get to decisions and actions faster.
To learn more, watch our recent IoT webinar with Jim McHugh, vice president and general manager of DGX Systems at NVIDIA, in which we talked about how GPUs accelerate IoT workloads.
If you’re in New York next week, join Kinetica at the O’Reilly AI conference, booth 20, where you can learn more about how GPU-accelerated databases is transforming virtually every vertical.