Top 5 Benefits OF Using Muvi Live Paywall For Live Streams
Rancher Meetup Deep Dive#01LT English Version
1. SCORER On Rancher
~ Make Computer Vision Easy ~
Image Analysis Computing Cluster
on
Rancher & Kubernetes
2. IP Camera
USB
Camera
IP Camera
( RTMPS Available)
Smartphone
Camera
AI
Log
Gateway
SCORER Edge
Input Analysis Output
Different AIs(Container)
Backend/ Computing Cluster
UI/App
Fully Managed
Cloud Environment
AWS S3 AWS S3
SCORER covers Image Capture to Analysis
Cloud API
AWS
Rekognition
Azure
Cognitive Service
SCORER
Video Collection System
View / Operation / Analysis
Data transfer
User
Environment
:SCORER Coverage
3. Why Rancher?
● AI is one of the components of SCORER Cloud
● All Image analysis AIs are containerized
● Each function(getting, playing, storing video) is also containerized
● Lot of users call analysis request so the system has to manage many
batch-job type containers
SCORER Cloud needs flexible container orchestration system
4. The issues with Rancher 1.6
● System Flexibility/Reliability
○ We chose Rancher 1.6 for the previous system. It used Rancher CLI(Cattle) to handle
containers but it had issues with flexibility/reliability.
● Host Resource Management
○ The system needed to choose between CPU/GPU hosts to deploy tasks based on the analysis
request. It was not impossible but difficult to implement the selection with Rancher 1.6.
We decided not to enhance the Rancher 1.6 system
but instead move to Rancher 2.x (with Kubernetes)
5. System Architecture with Rancher 2.x
● CPU/GPU nodes are mixed in a cluster
● Staging/Production environments are integrated within a
cluster
● Reduced EC2 instances for Rancher-Manager
8. Physical Configuration on AWS
Analysis nodes are
integrated by Rancher
Manager. The nodes are
distributed to multi
availability zone for
resilience.
The configuration follows the
official manual by Rancher
for High Availability
Configuration
9. Future:Multi Cloud/On-Premises mixed cluster
Rancher 2.x can handle multiple Kubernetes cluster.
It makes it possible to combine computing resources flexibly based on merit.
On-Premises
10. Summary
● We moved our computing cluster from Rancher 1.6 to Rancher 2.x
● It makes it possible to scale the system with on-premise cluster.
(Cloud GPUs are very very expensive if you want to use for an extended
period of time. )
Any opinions / suggestions are
more than welcome!!