Like a data center at the edge (and 73% cheaper)

Within 3 weeks of the COVID-19 pandemic, Camio launched social distancing and mask detection globally to help essential service companies remain open safely. Cost was secondary to the speed of deployment, since life safety and business continuity were at risk. The Cloud enabled that quick launch, because the new Camio AI models could run on GPUs in the cloud instantly without waiting for anything to be deployed at the edge.

Instant cloud deployments of GPU-heavy workloads were justified by life safety and business continuity.

But then retailers asked Camio to apply its same technology to in-store conversion analytics that couldn’t justify the same compute costs; continuous tracking and inference relied on server instances and GPUs in data centers that could cost more than human review. So the Camio team moved these workloads to the edge by running them in a Camio Flex Kubernetes cluster that reduced the cost by 73%.

In-store conversion analytics sounds simple but actually involves the orchestration of a lot of different services:

• human detection
• human pose detection
• human tracking
• dwell times
• human engagement
• object size calculation (adult/child)
• virtual turnstile counting
• classifiers for uniforms/logos/text
• BigQuery streaming for SQL data access
• query-based notifications

The reporting requirements made the orchestration of the workloads and instant delivery of new AI models just as important as raw compute power.

Camio Flex cluster at the edge

One of the fantastic things about the Kubernetes ecosystem is that big tech cloud providers have invested billions of dollars to enable sophisticated datacenter workloads to run on any computer. Camio Flex exploits this infrastructure to move workloads between the edge and cloud to fit within price performance and latency constraints of any targeted solution.

In the case of retail in-store conversion analytics, both uplink bandwidth and costs were severely constrained. So Camio, with support from Qualcomm Technologies, is making it simple to use the edge compute power of Qualcomm Inc.’s SoC (System on a Chip) by running services like the ones mentioned above in containers. The resulting system runs 73% cheaper than the cloud version and uses 98.6% less bandwidth.

Here are the inputs to the calculated cost savings of the differences between cloud and edge deployments:

Beyond 98% reduction in uplink bandwidth, Camio Flex cut costs 73%.

Camio orchestrates general purpose computing tasks and queues with Kubernetes. Edge AI deployments using Camio Flex enable device manufacturers and solution providers to run sophisticated computer vision applications for 1/4 the cost of running on cloud.