Network Service Mesh
networkservicemesh.io
- Free plan
- Open source
- Linux
- API
- Self-hosted
What Network Service Mesh is
Network Service Mesh is an open-source project for connecting workloads across clusters and clouds through a hybrid, multi-cloud IP service mesh. Each workload can connect independently to zero or more Network Services, wherever it runs. The project says it works with an existing CNI and needs no changes to Kubernetes or the workloads. Network Services can provide connectivity, security, and observability at Layer 3 and above, carrying IP or Ethernet payloads. Clients may be Kubernetes pods, virtual machines, or physical servers. They authenticate independently, currently using SPIFFE IDs, and must be authorized before attaching to a service. Clients use gRPC to request or close virtual wires and can monitor connection status through a streaming gRPC call. Endpoint selection supports composition, selective composition, topology-aware selection, and scale from zero. A service may also provide DNS scoped to specified domains or subdomains for connected pods. Its architecture describes use with Istio, Linkerd, Consul, and Kuma. The project is a CNCF sandbox project and its listed plan is free.
Who it suits
Network Service Mesh suits teams connecting workloads across Kubernetes clusters and clouds, including environments with virtual machines or physical servers. It expects operators to have Docker, Go, a Kubernetes cluster, and kubectl available; most examples require at least two worker nodes.
Good to know, and what to watch
- Connects workloads across clusters and clouds.
- Works with an existing CNI, according to the project.
- Supports IP and Ethernet payloads.
- Clients can monitor connection status through gRPC.
- Examples cover OpenTelemetry, OPA, IPv6, and DNS.
- Run guide prerequisites include Docker and Go.
- A Kubernetes cluster and kubectl are prerequisites.
- Most examples require at least two worker nodes.
- Clients need authorization to attach to a service.
Plans
Every plan and priceNetwork Service Mesh
Open source
Getting started
- Open the Network Service Mesh website
- Prepare Docker, Go, a Kubernetes cluster, and kubectl
- Use a supported environment such as Kubernetes, kind, minikube, microk8s, k3s, GKE, Azure, AWS, or bare metal
- Deploy the project and connect clients through its gRPC operations API
The Club's review
Network Service Mesh is a free, open-source networking layer for teams connecting workloads across Kubernetes clusters and cloud environments. It suits platform engineers managing mixed infrastructure who need each pod, virtual machine, or server to obtain its own network services. The trade-off is operational: it is powerful and adaptable, but its Kubernetes-centered prerequisites are substantial for anyone wanting a turnkey network.
Overview
Connectivity is requested per workload rather than fixed to a cluster. A client can attach independently to zero or more Network Services wherever it runs, while those services supply Layer 3-and-above connectivity, security, or observability. IP and Ethernet payloads are supported, making the model relevant to both conventional application traffic and workloads needing broader network semantics.
Network Service Mesh is designed to work with an existing CNI and requires no changes to Kubernetes or to application workloads. Authentication is independent for each client, currently using SPIFFE IDs, and authorization is required before attachment. That separation helps enforce which workloads may consume a service, although it also demands a capable identity and policy operation. The project is a Cloud Native Computing Foundation sandbox project.
Key features
Per-workload connections
Clients use gRPC to request and close virtual wires and can monitor connection state through a streaming gRPC call. This is a practical foundation for automation and controllers, rather than a purely static network configuration. It is best for teams comfortable integrating APIs into platform workflows.
Routing and service composition
Endpoint selection supports composition, selective composition, topology-aware selection, and scale from zero. These controls can keep traffic near the right topology, combine services deliberately, and avoid reserving capacity for idle workloads. They add useful flexibility for multi-cluster designs, but require thoughtful service architecture and operations.
Network services and integrations
A Network Service can provide DNS limited to specified domains or subdomains for connected pods. Published examples cover OpenTelemetry, OPA, IPv6, dual stack, Jaeger, SR-IOV, and Open vSwitch. The architecture also describes using services alongside Istio, Linkerd, Consul, or Kuma, so organizations can extend an existing service-mesh approach instead of treating this as a replacement for every mesh function.
Pricing
Network Service Mesh costs nothing under the Network Service Mesh plan. It is an open-source CNCF sandbox project, with no paid tier or subscription term described. The free status is attractive for experimentation and production deployments that can support the project themselves; it does not remove the engineering and infrastructure work involved in operating it.
Platforms
The project is delivered through an API on Linux and is self-hosted. Supported environments include Kubernetes, kind, minikube, microk8s, k3s, GKE, Azure, AWS, and bare metal. The run guide requires Docker, Go, a Kubernetes cluster, and kubectl, and most examples require at least two worker nodes. Those requirements make it a poor fit for a single-node or lightweight deployment.
Who it's for
Choose Network Service Mesh when platform or network teams need policy-controlled, per-workload connectivity spanning clusters, clouds, and non-container clients. Its support for pods, virtual machines, and physical servers is especially useful in transitional or hybrid estates, while encryption in transit, network policies, and multi-cluster networking address core operational concerns.
Look elsewhere if the priority is a minimal CNI, a managed service, or an installation with few prerequisites. You will need Kubernetes-oriented operational skills, identity authorization based on SPIFFE IDs, and capacity to run the supporting components and integrations.
Pros and cons
Pros
- Per-workload attachment: each client can independently connect to multiple services regardless of location.
- Broad workload coverage: pods, virtual machines, and physical servers can act as clients.
- Flexible traffic model: IP and Ethernet payloads, topology-aware selection, and service composition support complex designs.
- Policy and security controls: independent SPIFFE authentication, authorization, encryption in transit, and network policies are built into the stated model.
- Open-source cost: the plan is free with no subscription charge.
Cons
- Heavy prerequisites: Docker, Go, Kubernetes, kubectl, and often two worker nodes raise the deployment bar.
- Self-managed operations: the free model leaves teams responsible for running and integrating the system.
- Sandbox status: organizations may prefer a more mature project for critical networking.
- Complexity at scale: composition, authorization, and topology choices reward specialist operators rather than casual users.
Alternatives
Canal is a free, open-source option for teams wanting a simpler Linux, macOS, API, or self-hosted networking project.
Calico Open Source fits organizations seeking free networking with unlimited clusters and community-driven maintenance across Linux, Windows, and self-hosted deployments.
DANM is a free Kubernetes network-management choice for Linux and self-hosted environments.
Kube-router offers a free, open-source Kubernetes networking solution on Linux and self-hosted systems.
Spiderpool is a free Apache-licensed Kubernetes networking project with API, Linux, and self-hosted support.
Terway is another free, open-source Linux and self-hosted option; trunking is unavailable in self-hosted clusters.
VMware Workstation Pro is free without a subscription or license key for supported 64-bit Windows or Linux hosts, but it serves a different virtualization use case.
Aviatrix Kubernetes Firewall is the paid alternative with a 30-day free trial for teams evaluating a commercial firewall product.
For broader comparisons, see Container Networking Software and Service Mesh Platforms.
Verdict
Network Service Mesh is a strong choice for platform teams that need programmable, identity-aware network services across clusters, clouds, and heterogeneous workloads. Its main reason to choose it is the per-workload model combined with rich routing and payload options. Look elsewhere if you cannot meet the Kubernetes and multi-node operating requirements or want a managed, low-maintenance network layer.
The verdict
Network Service Mesh provides a way to connect individual workloads to network services across clusters and clouds. Its free, open-source status may appeal to teams prepared to meet the deployment prerequisites.
Questions people ask
How much does Network Service Mesh cost?
The listed Network Service Mesh plan is free. The project is open source.
Which clients can connect?
A client can be a Kubernetes pod, virtual machine, or physical server.
Does it require Kubernetes changes?
The project says it requires no changes to Kubernetes or workloads and works with an existing CNI.
What platforms are supported?
Listed targets include Kubernetes, kind, minikube, microk8s, k3s, GKE, Azure, AWS, and bare metal.
Does it provide network security features?
Network Services can provide security, and the listing includes network policies and encryption in transit.
What the maker publishes
- Advanced routing
- Endpoint selection supports composition, selective composition, topology-aware selection, and scale from zero. source
- Cloud integrations
- The run guide lists GKE, Azure and AWS integrations for system tests, and release 1.5.0 reports testing on GKE, AKS and AWS. source
- CNCF acceptance
- CNCF records Network Service Mesh as accepted at Sandbox maturity on April 11, 2019. source
- CNI compatibility
- NSM keeps the existing Kubernetes CNI for intra-cluster networking and does not require replacing it. source
- Connection monitoring
- The MonitorConnection gRPC service streams connection events for selected connections. source
- Core capabilities
- The project describes L3 zero trust, per-workload granularity, and per-workload connectivity, security and observability. source
- Deployment requirements
- The run guide lists Docker, Go, a Kubernetes cluster, and kubectl as prerequisites; it says most examples require at least two worker nodes. source
- DNS
- A Network Service may provide DNS scoped to specified domains or subdomains for connected pods. source
- DNS and IPAM
- NSM can provide per-network-service DNS, inject interfaces into pods, assign endpoint-provided addresses and add optional routes while avoiding cluster-prefix collisions. source
- Documentation depth
- The examples catalog includes inter-domain, Istio, OpenTelemetry, OPA, IPv6, dual-stack, Jaeger, DNS, webhook, SR-IOV, OVS, resilience and topology-aware examples. source
- Examples and integrations
- The project lists examples for OpenTelemetry, OPA, IPv6, dual stack, Jaeger, DNS, SR-IOV, and Open vSwitch. source
- Forwarder backends
- The reference components include VPP and Open vSwitch forwarders for creating network interfaces. source
- Governance
- Project governance prioritizes community over product or company and says contributors participate as individuals. source
- Identity and authorization
- Clients are independently authenticated, currently by SPIFFE ID, and must be authorized to attach to a Network Service. source
- Independent event description
- A 2020 NSMCon prospectus characterized NSM as community-driven, CNCF sandbox software extending IP reachability to multi-cluster, legacy, on-premises and public-cloud workloads. source
- Install prerequisites
- The run guide lists Docker, Go, kubectl and a Kubernetes cluster; kind, minikube, microk8s and k3s are suggested cluster options. source
- Inter-domain registry
- Registry domains can be independent, and clients and endpoints may connect across domains subject to policy. source
- Kubernetes admission
- The Kubernetes admission webhook injects NSM clients into annotated pods or into pods in annotated namespaces. source
- Kubernetes fit
- The project says it requires no changes to Kubernetes or workloads and works with an existing CNI. source
- License
- The public Network Service Mesh GitHub repositories shown are licensed Apache-2.0. source
- Limit: latency and ingress concerns
- The same analysis reports that service-mesh users can encounter added communication latency and may need separate Kubernetes ingress controllers for external traffic. source
- Limit: operational complexity
- TechTarget's independent service-mesh analysis warns that mesh infrastructure adds data and control planes, increases container-environment complexity and can add compute overhead. source
- Network Service Manager
- NSMgr discovers services and endpoints, processes client requests, stores valid connections and can optionally act as a registry. source
- Network service payloads
- Network Services can carry IP or Ethernet payloads, with IP as the default when unspecified. source
Checked against the maker's pages on 15 Oct 2026.