Skip to main content
You are here: Further information - Architecture and compatibility

Networking

Public Network (Services and Access)

The Public Network from the GPUaaS Node is used both for GPU access and for exposing user services.

  • Service Exposure (20000–29999 TCP/UDP)
    Services can be exposed using this port range to allow users to make their running applications publicly accessible.

  • GPUaaS Instance Access (30000–32767 TCP/UDP)
    Users access their assigned GPUaaS instances via SSH within this port range.

Connection between the Admin/User Panel and GPUaaS Nodes requires a port for SSH (non-standard, or standard), any port you like, as we can specify this from the Admin Panel.

Private (Node Network/Management)

Please avoid 10.96.0.0/12 when assigning a private network to your nodes, to avoid overlapping with the Kubernetes Cluster Network.

Controller + Worker (first node you'll install)

Inbound

TCP/6443, API server

TCP/10250, kubelet (kubelet)

TCP/10256, kube-proxy

TCP/2379–2380, etcd

TCP/179, Calico BGP (internal)

Outbound

TCP/10250, to worker kubelets

TCP/179, to Calico BGP peers (internal)

TCP/22, to the admin/user panel

GPUaaS Nodes (workers any other node after initialisation)

Inbound

TCP/10250, kubelet

TCP/10256, kube-proxy

TCP/179, Calico BGP (internal)

Outbound

TCP/6443, to API server

TCP/179, to Calico BGP (internal)