Recently, NVIDIA launched Morpheus, a new open AI application framework powered by NVIDIA GPUs combined with NVIDIA®
With the launch of Morpheus, NVIDIA is bringing the powers of artificial intelligence to cybersecurity. The highly optimized AI pipeline and pre-trained AI capabilities of Morpheus allow developers to inspect all IP traffic across the data centre fabric instantaneously.
How does Morpheus work
Morpheus provides a framework to carry out real-time inference across large amounts of telemetry. Thanks to the use of GPUs across the workflow – from data ingestion, into pre-processing, to inference, through post-processing. GPUs keeping all the data end-to-end eliminates expensive serialization and deserialization actions. GPUs offer massive parallelization to move thickets of data through the pipeline. By dividing the data and the actions into manageable batches and executing them concurrently, Morpheus keeps up with data flowing from heterogeneous sources.
Morpheus can perform real-time inference across massive amounts of cybersecurity data and receive network telemetry from the NVIDIA BlueField-2 Smart NIC. Telemetry agents running on the NVIDIA DPU channel data to Morpheus. For sensitive information detection and phishing detection, this is often full packet data. However, this isn’t a one-way stream. Morpheus generates actions from raw inference results that are routed back to the NIC. The actions allow continuous, real-time, and variable feedback to the NIC that can impact policies, rewrite rules, adjust sensing, etc.
Morpheus is built on technologies, including RAPIDS, Cyber Log Accelerators (CLX), Triton Inference Server, TensorRT, and cuStreamz. A pub/sub model (currently Kafka) is leveraged to send data to and results from the inference pipeline. The technologies work together to address all parts of the cybersecurity workflow.
Morpheus is a perfect solution for gathering real-time network data from any DPU-powered server in the data centre. NVIDIA BlueField DPU offloads accelerate and isolate mission-critical data centre infrastructure functions. By incorporating the framework into a third-party cybersecurity offering, communication networks can benefit from the world’s best AI computing. Additionally, BlueField DPU goes beyond static security logging to incorporate a sophisticated dynamic real-time telemetry model. Developers can use the models through common deep learning frameworks like Caffe or Theano. The deep learning and machine learning models can be used in a Morpheus pipeline to detect leaked sensitive information, shady profiles, and phishing attempts.
NVIDIA collaborates with leading platforms and technology partners to optimize their data centre solutions for the NVIDIA Morpheus AI platform.
“Defending complex and evolving environments requires constant visibility,” said Adam Mishler, chief information security officer at Best Buy. “Providing real-time, dynamic network maps will help identify areas where we can further strengthen our posture and serve as a foundation for enhancing ML-based anomaly detection. The NVIDIA Morpheus framework helps provide a flexible and scalable platform for anomaly detection capable of adapting with the ever-changing cyber-threat landscape.”
Spunk’s senior vice president and Chief technology officer said with the Morpheus framework, the company can rapidly prototype new capabilities and offload compute-intensive tasks to GPUs to benefit its customers.
Cybersecurity breaches have gone up drastically, especially post-pandemic. Despite many advances in cybersecurity, it remains primarily reactive. Meaning, most of the remediation happens after the fact. The average time to scope out a breach is around 200 days, while the fixing takes 70 days. Frameworks like Morpheus can save significant downtime and, by extension, a lot of money and resources. Moreover, Morpheus is extensible. Developers can deploy their models using this framework and build on the work they have already invested in.