Client:
A leading Thai jewelry manufacturer.
The Challenge
A leading Thai jewelry manufacturer wanted to optimize production by analyzing worker activity across 200-250 workstations. However, processing 2TB of daily video data for real-time insights presented a massive computational challenge. They needed a secure solution for extracting key details (feature extraction & classification) to identify inefficiencies and make data-driven decisions.
2-2.5 TB Daily Video
Processing a staggering 2TB of video data daily from each workstation poses a significant computational challenge.
AI-powered Insights
The client needs to extract key details from video data (feature extraction and classification) to identify patterns and inefficiencies in worker performance.
Secure Video Data
Safeguarding sensitive worker activity and production data is crucial (compliance & security).
Cost optimization
The initial Network Video Recorder (NVR) solution proved to be quite expensive to operate. With 200 workstations each recording 9 hours of video daily, this amounted to approximately 1800 hours of video data that needed to be processed overnight. The client required an optimized solution for handling such large volumes of data in a cost-effective manner.
Our Solution
We calculated that the NVR solution proposed by the developers is quite expensive. It would require a lot of CPU to make the project financially viable
We proposed a significant cost optimization for this AI vision processing infrastructure by leveraging Azure Spot Virtual Machines (VMs). Spot VMs offer unutilized Azure compute capacity at deep discounts compared to traditional on-demand pricing.
This makes them ideal for workloads like video processing that can tolerate short interruptions.
Infrastructure Costs Estimation with Spot Instances
Here’s a breakdown of the estimated cost savings:
- Original Cost (On-demand pricing): $5,363 per month (as shown in the table)
- VMSS Cost with Spot VMs: $430 per month (significant reduction from $4,700)
- Total Estimated Cost with Spot VMs: $1,100 per month
Projected Savings: $4,263 per month
By utilizing Spot VMs for the VMSS instances, we can achieve a substantial cost reduction of approximately 81% compared to the initial estimate. This translates to savings of $4,263 per month.
Additional Benefits:
- Scalability: Azure’s cloud infrastructure allows scaling Spot VMs up or down as needed, ensuring efficient resource utilization.
- Performance: Spot VMs deliver high performance ideal for demanding workloads like video processing.
Considerations: While rare, Spot VMs can be interrupted when needed by Microsoft for other purposes. But our DevOps architect designed the architecture to handle these interruptions gracefully, ensuring minimal disruption to client’s AI vision processing tasks.
Using Azure’s scalable cloud infrastructure, we built a secure AI vision system. This system seamlessly processed the massive video data, extracted crucial insights, and provided real-time feedback for optimizing production processes.
- Designed a scalable and secure cloud architecture on Azure to handle the massive data volume and real-time processing requirements.
- Developed pipelines for efficient data ingestion, storage, and processing
Deployment diagram
