Networking

The networking layer controls how traffic flows between workloads, hosts, sites, and users. It enforces segmentation by default, encrypts traffic between services, and provides the communication fabric that every other layer depends on.

Networking is the layer where most security decisions are made — and where most security mistakes are made. The Nyxever approach is defense in depth: segment traffic so that a compromised workload cannot reach unrelated services, encrypt everything even on internal networks, and authenticate services to each other rather than relying on network position as a trust signal.

Flat networks — where every host can reach every other host — are convenient but give a compromised host a path to everything. Nyxever segments the network so that every workload gets the network access it needs and nothing more.

Segmentation

Overview

Traffic segmentation divides the network into zones that have different trust levels and communication rules. Workloads can only communicate with other workloads in zones they have explicit access to. A database server in a storage zone should not be reachable from a public-facing web server — even though they're on the same physical infrastructure.

Segmentation limits blast radius: when a workload is compromised, what an attacker can reach from it is constrained. Without segmentation, a compromised container has the same network access as the hypervisor host. With proper segmentation, it has access to the specific services it needs to function.

⚠  Segmentation zone model and access control matrix documentation in progress.

Architecture

The design approach is 802.1Q VLAN segmentation within a site, with overlay networks (VXLAN) considered for segments that must span multiple physical locations. Traffic between segments is routed, and explicit ACLs at the routing layer control what can talk to what — each permitted flow between segments is an explicit rule.

Typical zones: management (out-of-band host management), storage (Ceph replication and client traffic), cluster (Proxmox VE cluster communication via Corosync), workload (VM/container data plane), VPN (remote access termination), and services (any externally-accessible services). These are starting points, not a required topology.

Inside Kubernetes, segmentation continues at the pod level: NetworkPolicy resources define which pods may talk to which, and are enforced by the cluster's CNI implementation. Pod-level policy complements the VLAN zones; it does not replace them.

⚠  VLAN numbering scheme, ACL templates, and inter-zone routing documentation in progress.

Technologies

Layer 2 segmentation is built on open standards: 802.1Q VLAN tagging, supported by managed switches generally and natively by the Linux kernel's network stack. For segments that need to span multiple physical sites or that need to be decoupled from the physical VLAN topology, VXLAN overlays extend L2 segments over L3 infrastructure.

Under evaluation: whether multi-site segments use static VXLAN tunnels or a BGP EVPN control plane (Proxmox VE's SDN feature supports both), and the switching and routing hardware, which will be selected on standards support, operability, and cost.

VLANs

802.1Q VLANs are the standard mechanism for segmenting Layer 2 broadcast domains. A managed switch port is either an access port (carries traffic for one VLAN, untagged) or a trunk port (carries traffic for multiple VLANs, tagged with the VLAN ID in each frame's 802.1Q header). Hypervisor hosts connect to switch trunk ports, so each VM can be placed on its VLAN without requiring a physical port per VLAN.

⚠  VLAN configuration guide for Proxmox VE and Linux bridges in progress.

VXLANs

VXLAN (Virtual Extensible LAN) encapsulates Layer 2 frames inside UDP packets, enabling L2 segments to span L3 network boundaries. Each VXLAN segment is identified by a 24-bit VNI (VXLAN Network Identifier), providing up to 16 million segment IDs — compared to 4094 for 802.1Q VLANs. VXLAN is primarily relevant for multi-site deployments where VLAN-level connectivity between sites is impractical.

⚠  VXLAN overlay configuration and multi-site networking documentation in progress.

VPN

Overview

VPNs create encrypted tunnels between endpoints — between two sites (site-to-site), between a remote user and the infrastructure (remote access), or between a service and an external dependency. VPNs are not a substitute for network segmentation — they provide encryption and authentication for a connection, not access control inside the connected networks.

In Nyxever, VPNs serve two primary uses: connecting physically separate sites so that their Ceph clusters and Proxmox VE clusters can communicate securely over untrusted links, and providing authenticated remote access for operators without exposing management interfaces to the internet.

⚠  VPN design guide and site-to-site topology documentation in progress.

Architecture

A VPN tunnel authenticates both endpoints and encrypts all traffic between them. Modern VPN protocols use public key cryptography for the handshake — each endpoint has a key pair, and the connection is established only if both endpoints can prove possession of their private key. Traffic is then encrypted with a symmetric cipher derived from the handshake, rekeyed periodically to provide forward secrecy.

⚠  Key distribution, certificate vs pre-shared key authentication, and PKI integration documentation in progress.

Technologies

The VPN technology is not yet decided. The requirements: strong, modern cryptography with forward secrecy; mutual authentication of both endpoints; a key or credential revocation process that is practical at the scale we operate; support for routed site-to-site links; client support on the platforms operators use; and a small, understandable configuration surface.

Under evaluation: two standard tunnel protocols, described below. The choice may differ between site-to-site links and operator remote access.

IPsec / IKEv2

IPsec is the IETF standard suite for encrypting IP traffic, with IKEv2 (RFC 7296) handling key exchange. It is widely supported by network hardware and operating systems, and can authenticate peers with X.509 certificates — which ties revocation and rotation into a PKI. The tradeoff is a large configuration surface with many cipher and mode options that must be constrained deliberately.

WireGuard protocol

WireGuard is a VPN protocol with a fixed modern cipher suite: Curve25519 for key exchange, ChaCha20-Poly1305 for encryption, and BLAKE2s for hashing. It is included in the mainline Linux kernel and has implementations for other platforms. Its small codebase and lack of cipher negotiation make it comparatively easy to audit and reason about.

Peers authenticate with static public keys rather than certificates. That removes the need for a PKI, but revocation means removing a key from every peer that trusts it — workable for a few long-lived infrastructure links, harder for many users.

⚠  VPN protocol comparison and decision record in progress.

Service Mesh

Overview

A service mesh handles east-west traffic — communication between services within the infrastructure. The goal is to authenticate every service-to-service connection and encrypt it, regardless of the network segment they're on. A service mesh implements zero-trust networking: network position does not grant trust; cryptographic identity does.

Without a service mesh, a service that can reach another service on the same VLAN can impersonate any other service — there is no authentication at the service level. With mTLS (mutual TLS), every connection requires both sides to present a valid certificate, and that certificate carries the service's identity. Compromising one service does not give you credentials to impersonate another.

⚠  Service mesh topology and certificate distribution documentation in progress.

Architecture

The requirement is that every service-to-service connection is encrypted with TLS 1.3 and mutually authenticated, using short-lived certificates issued by an internal X.509 PKI and rotated automatically (for example via ACME, RFC 8555). Authorization is then based on the identity in the certificate, not on source address.

Under evaluation: two ways to meet it. Services can terminate mTLS themselves, each obtaining its own certificate and checking client identities against an allowlist — fewer moving parts, more work per service. Or, for workloads on Kubernetes, a service mesh can handle mTLS transparently with sidecar or node-level proxies — consistent enforcement, but an additional runtime to operate and trust. Workload identity standards such as SPIFFE/SVID are part of the evaluation.

⚠  Certificate issuance, rotation automation, and service identity patterns documentation in progress.

Technologies

mTLS

Mutual TLS (mTLS) extends standard TLS to require the client to present a certificate as well as the server. Both sides authenticate each other. The certificate carries the service's identity — typically encoded in the Subject Alternative Name (SAN) as a URI or DNS name. The receiving service verifies that the connecting client's certificate was issued by a trusted CA and that the identity in the certificate is authorized to make the request.

⚠  mTLS configuration examples and certificate identity patterns in progress.

DNS

Overview

Internal DNS resolves service names to addresses within the infrastructure without exposing that information externally. Split-horizon DNS serves different answers to internal and external resolvers — an internal service might have a real IP on the internal DNS, while external DNS for the same name returns a public load balancer or simply doesn't exist.

DNS is also a service discovery mechanism. Rather than hardcoding IP addresses in configuration, services use DNS names that can be updated when addresses change. In container and VM environments where IP addresses are frequently reassigned, DNS-based service discovery is essential for stable addressing.

⚠  Internal DNS setup, split-horizon configuration, and service discovery documentation in progress.

Technologies

The DNS software is not yet decided. Requirements: separate authoritative and recursive roles, split-horizon views for internal and external names, DNSSEC signing and validation, dynamic updates (RFC 2136) or an API for automation, and clean integration with Kubernetes cluster DNS for service discovery inside the cluster.

⚠  DNS server evaluation and decision record in progress.