Problem to Solve
User story
As a cluster admin, I want to deploy an HA cluster.
As an application developer, deploying network intensive workload like a visioconferencing tool, I want to be able to have limitless bandwidth.
Context
From what we understand from outscale internal network:
- each VPC comes with a firewall VM
- this firewall VM has a limit of ~300k packets/second
In this document, we read that to be truly HA, the best is to deploy 2 VPCs, and peer them together.
https://docs.outscale.com/fr/userguide/Designing-a-Resilient-Infrastructure.html#_creating_a_multi_subregion_architecture
For our visio usecase, the current outscale recommendation, and our current implementation is to:
- deploy one kubernetes cluster with cluster api in a main VPC
- for the visio part, we deploy one VPC/VM that we peer with the main VPC (we currently have 10)
Without the kubernetes overhead, I'm curious to hear how can I make
Proposed Solution
Not sure about the best solution, just giving some ideas.
For the HA use case
we could imagine the following:
If we deploy the cluster in multiAZ, it means that the end user wants HA.
In this case, we could deploy 3 VPCs, one per AZ and peer all of them together.
For the network scaling use case
We could imagine to have a new param in oscCluster:
numberOfAdditionalVPC: 10
And then in the MD object we could specify the VPC id where to schedule (or think of a "repulsion" of VM based on the VPC label ?)
Additional Context
The LB is also only one VM. We would need to figure out a way to also make this HA.
I'm curious how much uptime we would gain with this added complexity, compared to downtime risks based on this complexity.
Environment Details
Problem to Solve
User story
As a cluster admin, I want to deploy an HA cluster.
As an application developer, deploying network intensive workload like a visioconferencing tool, I want to be able to have limitless bandwidth.
Context
From what we understand from outscale internal network:
In this document, we read that to be truly HA, the best is to deploy 2 VPCs, and peer them together.
https://docs.outscale.com/fr/userguide/Designing-a-Resilient-Infrastructure.html#_creating_a_multi_subregion_architecture
For our visio usecase, the current outscale recommendation, and our current implementation is to:
Without the kubernetes overhead, I'm curious to hear how can I make
Proposed Solution
Not sure about the best solution, just giving some ideas.
For the HA use case
we could imagine the following:
If we deploy the cluster in multiAZ, it means that the end user wants HA.
In this case, we could deploy 3 VPCs, one per AZ and peer all of them together.
For the network scaling use case
We could imagine to have a new param in oscCluster:
numberOfAdditionalVPC: 10
And then in the MD object we could specify the VPC id where to schedule (or think of a "repulsion" of VM based on the VPC label ?)
Additional Context
The LB is also only one VM. We would need to figure out a way to also make this HA.
I'm curious how much uptime we would gain with this added complexity, compared to downtime risks based on this complexity.
Environment Details