Employing 1,000 people, CELEPAR (Companhia de Tecnologia da Informação e Comunicação do Paraná) develops software solutions for the state of Paraná, Brazil. The organization builds and manages IT systems that support key public services, including healthcare, policing and education. CELEPAR operates its own resilient and highly secure Tier III data center infrastructure, enabling it to protect sensitive data on behalf of users and residents throughout the state.
At-a-Glance
To empower government agencies in the Brazilian state of Paraná to spend less time on repetitive manual tasks and more time on front-line services, CELEPAR used the SUSE Rancher Prime platform to deploy a first-of-its-kind biometrics service. Built on GPU-accelerated microservices on Kubernetes, the new AI and machine learning solution helps government agencies accelerate their verification and identification processes, giving residents faster access to public services.
AI-Powered Facial Recognition at Scale with SUSE Rancher Prime
“With SUSE Rancher Prime, we can ensure that we use our powerful NVIDIA hardware to its full potential. The SUSE stack is helping us to maximize the business value of our investment in NVIDIA A100 Tensor Core GPUs.”
The journey to containers
CELEPAR strives to build digital solutions that enrich and improve the lives of the millions of people living in Paraná. The company is always looking for ways to use cutting-edge technology to support new and better public services — and saw advances in artificial intelligence (AI) and machine learning (ML) as powerful opportunities for innovation.
Danilo Souto, Coordinator of Strategic Data at CELEPAR, says: “Identity verification is a crucial first step before residents access public services. Historically, government agencies have relied on people to perform these checks manually, which consumed an enormous amount of time.”
To streamline the process without compromising security, CELEPAR decided to build an intelligent facial biometrics platform. The aim was to use video or static images to automatically validate identities, saving valuable time both for residents and government employees.
Building the new biometrics platform presented several technical challenges. CELEPAR’s existing IT infrastructure couldn’t deliver the compute performance to perform model inference at the required speed and scale. Additionally, although the environment was built on a modern containerized architecture, the existing container management platform lacked support for GPU-accelerated workloads: a critical requirement for the new AI and ML workloads.
Rafael Rudnik, System Analyst at CELEPAR, continues: “When we started the biometrics project, there was no official support for NVIDIA GPUs in our existing container orchestration solution. We also felt that implementing AI and ML workloads — which impose high memory and parallel processing demands — might cause instability for our existing mission-critical production microservices.”
To minimize risk, CELEPAR created a completely new containerized environment for the new AI and ML models. Due to the sensitive nature of the data to be processed, the company knew that an on-premises environment would be crucial to meet its stringent legal and data sovereignty obligations.
Why SUSE Rancher Prime?
Next, CELEPAR set out to find a container management platform with native support for GPU-accelerated workloads — and chose SUSE Rancher Prime as the cornerstone of its innovative biometrics platform.
“One of the key reasons for selecting SUSE Rancher Prime was its proven and officially supported integration with our NVIDIA GPUs,” explains Souto.
CELEPAR’s new environment is based on four Dell PowerEdge R7525 server nodes powered by AMD EPYC 7413 processors. Each node is equipped with 9 TB of NVMe storage, and an NVIDIA A100 80GB Tensor Core GPU delivers optimal performance for AI and ML workloads.
With SUSE Rancher Prime, CELEPAR can use NVIDIA Multi-Instance GPU (MIG) technology to partition each NVIDIA A100 Tensor Core GPU into smaller, isolated instances. This allows the company to run multiple inference models efficiently and allocate GPU resources precisely as needed.
“We rely on SUSE solutions to help us extract maximum performance from our infrastructure,” says Rudnik. “All our nodes run SUSE Linux Enterprise Server, and we use SUSE Rancher Prime to orchestrate our Kubernetes environment and manage NVIDIA MIG. This combination provides a holistic and fully supported stack from the hardware level up to the applications, significantly speeding up incident resolution. It also means that we can leverage the performance and hardware capability of SUSE Linux Enterprise Server.”
Developed in-house and built on open source models, running on the TensorFlow stack, CELEPAR’s facial recognition service is provided through two APIs. The first API offers facial verification by comparing a picture of a subject with a secure embedding code of verified photos, while the second one is a validation API based on ICAO (International Civil Aviation Organization) patterns, to perform a complete facial match and validation.
The impact of SUSE solutions
Unlocking high performance for AI workloads
With SUSE Rancher Prime and SUSE Linux Enterprise Server, CELEPAR has unlocked the GPU-accelerated performance required to run its innovative biometrics workloads. The foundational stability and optimized performance provided by running nodes on SUSE Linux Enterprise Server have been crucial in supporting these demanding AI/ML workloads and ensuring the reliability of the overall solution. The new AI and ML environment is the first of its kind in Paraná, and is already handling a large volume of verification and identification requests from government agencies across the state.
“Our biometric service manages around 35,000 API calls a month, but we have a roadmap to scale up to more than 130,000 API calls a month as more agencies incorporate the solution into their workflows,” says Rudnik. “Without native GPU support from SUSE Rancher Prime, it would have been extremely difficult to reach this milestone.”
Building expertise in GPU-based Kubernetes
During the implementation, CELEPAR worked closely with local SUSE partner Pointer, which helped the project team gain invaluable hands-on skills. SUSE Rancher Prime’s flexibility enables CELEPAR to experiment with, deploy and manage AI applications with much greater freedom and agility than its existing container management platform.
“We went from having no experience with GPU-based Kubernetes to becoming proficient experts managing a complex, mission-critical environment,” confirms Souto. “We find the SUSE Rancher Prime user interface intuitive and easy to use, especially for tasks like managing pods, viewing logs and configuring resources.”
Improving operational cost-efficiency
The combination of SUSE Rancher Prime and NVIDIA GPU technology enhances the cost efficiency of CELEPAR’s new AI-powered service. By directly managing NVIDIA MIG technology in Rancher, the company can precisely partition and allocate its GPU resources.
Rudnik comments: “With SUSE Rancher Prime, we can ensure that we use our powerful NVIDIA hardware to its full potential. The SUSE stack is helping us to maximize the business value of our investment in NVIDIA A100 Tensor Core GPUs.”
Gaining peace of mind with 24/7 support
Backed by SUSE Rancher Prime Priority Support, CELEPAR has the peace of mind that rapid, enterprise-grade support from dedicated SUSE experts is always available across their entire SUSE stack, including SUSE Linux Enterprise Server. This truly holistic support spans every layer of their SUSE-powered container environment, from the SUSE Linux Enterprise Server operating system and the Kubernetes distribution to the Rancher Prime management platform, ensuring a single point of contact for comprehensive and faster incident resolution.
“Our biometric platform has been up and running for more than two years, and in all that time we’ve only needed to contact SUSE Support on two or three occasions,” says Rudnik. “Our SUSE Rancher Prime environment is very easy to manage, and we can perform most of the work ourselves or with assistance from Pointer. We see SUSE Rancher Prime Priority Support as our safety net; if anything goes wrong, we know we can count on SUSE.”
What’s next for CELEPAR?
At CELEPAR, innovation never stands still. The company is already planning the next evolution of its biometric service, which includes adopting the security- and compliance-focused RKE2 distribution. The organization is also exploring other solutions in the SUSE stack, including SUSE Security and SUSE Observability.
“We are currently building our own fine-tuned models to enhance the accuracy of our facial recognition technology, and we plan to use our SUSE Rancher Prime environment to train the new models,” concludes Rudnik. “By building on our partnership with SUSE, we see great potential to further enhance the security and manageability of our AI and ML infrastructure.”