highlights and lowlights of the daily business of network engineers

Download als PDF File

Purpose

These notes summarize the main iMaster NCE-Campus deployment models: when they are used, how high availability and geographic redundancy are implemented, and what the minimum requirements for CPU, RAM, disk, and NICs are.

Overview of Deployment Variants

1) Single-node system

• All product components run on a single node.
• Suitable for small environments and simple installations.
• PM-based deployment is recommended.
• Typical scale:
  • LAN only: up to 5,000 LAN-side NEs and up to 20,000 online users
  • LAN-WAN convergence: LAN-side NEs + WAN-side NEs x 5 <= 5,000 and up to 20,000 online users

Useful for small environments, lab setups, or scenarios where simplicity is more important than built-in cluster redundancy.

2) Minimum cluster

• 3 nodes.
• Controller/service, middleware, and big data analysis services are co-deployed on each node.
• PM-based deployment is recommended, but VM-based deployment is also possible.
• This is the typical smallest production-grade cluster variant.
• Typical scale:
  • LAN only: up to 30,000 LAN-side NEs and up to 100,000 online users
  • LAN-WAN convergence: up to 15,000 weighted devices and up to 50,000 online users

Useful when more availability than a single-node system is required, but a larger distributed architecture is not yet necessary.

3) Distributed cluster

• Starts at 6 nodes and scales in the standard designs to 9 or 17 nodes.
• VM-based deployment is recommended.
• Services are separated by role instead of being co-deployed on all nodes.
• Typical roles:
  • service/controller nodes
  • middleware nodes
  • big data analysis nodes

Typical scaling levels:

• 6-node distributed cluster:
  • up to 30,000 weighted devices
  • up to 100,000 online users
• 9-node distributed cluster:
  • up to 60,000 weighted devices
  • up to 300,000 online users
• 17-node distributed cluster:
  • up to 200,000 weighted devices, with WAN-side NEs <= 20,000
  • up to 700,000 online users

Useful when more scale, clearer role separation, and better performance isolation are required.

4) Multi-cluster system

• A global cluster is combined with multiple regional clusters.
• In the resource planning guide, this model is described as 3 global nodes plus up to 10 regional clusters.
• In the installation scenarios guide, one example shows 3 OMP nodes, 3 global nodes, and at least two regional distributed clusters with 9 or 17 nodes.
• Regional clusters operate like independent controller systems for their own region.
• Regional clusters cannot be single-node systems; they must be minimum or distributed clusters.

Useful for very large or geographically distributed organizations that want to centrally manage multiple regions.

Choosing the Right Variant

• Single-node for labs, PoCs, or small campus environments.
• Minimum cluster for production cluster deployments of medium size.
• Distributed cluster for large current or future growth and when roles should be cleanly separated.
• Multi-cluster when several regional deployments must be centrally aggregated.

Huawei recommendation from the installation scenarios document:

• LAN-only:
  • if <= 30,000 devices and no foreseeable expansion: minimum cluster
  • if > 30,000 devices or expansion is planned: distributed cluster
• LAN-WAN convergence:
  • if <= 15,000 weighted devices and no foreseeable expansion: minimum cluster
  • if > 15,000 weighted devices or expansion is planned: distributed cluster

HA, Redundancy, and DR

Cluster HA within one data center

• A minimum cluster already provides more availability than a single-node system because services are distributed across 3 nodes.
• A distributed cluster further increases resilience because service, middleware, and big data roles are separated across multiple nodes or VMs.
• In VM deployments, nodes of the same type must use anti-affinity.
• This means two VMs with the same role must not run on the same physical host.

This is important so that the failure of a single host does not cause several identical roles to fail at the same time.

Geographic redundancy / disaster recovery across multiple sites

• Huawei explicitly refers to a Geographic Redundancy System Installation for:
  • single-node DR
  • cluster DR
• An automatic DR solution with arbitration is recommended.
• The primary site, secondary site, and arbitration site must be deployed at different locations.
• Geographic redundancy should be understood as an active/standby design across separate locations.
• If the heartbeat between clusters fails, the arbitration service decides active/standby status.

Important: geographic redundancy is not simply „more nodes in one DC“, but a true multi-site DR design.

Arbitration node

• Used for the automatic DR solution with arbitration.
• According to Hedex, the arbitration service consists of ETCD and Monitor.
• Monitor checks connectivity between sites and writes the result into ETCD.
• Minimum requirements:
  • 1 node
  • 4 vCPUs
  • 16 GB RAM
  • 150 GB disk
  • 1 or 2 GE NICs

Minimum Resource Requirements

Single-node system

LAN-only single node

• 1 PM or 1 VM
• CPU:
  • PM: 40 threads, no overcommitment, 2.2 GHz+
  • VM: 40 vCPUs, 2.2 GHz+
• RAM:
  • 128 GB standard
  • 192 GB with IoT management
  • 256 GB if network configuration verification and runbook are installed together
• Disk:
  • 960 GB system disk
  • 1200 GB data disk
• NIC:
  • 1 or 2 GE

LAN-WAN single node

• 1 PM or 1 VM
• CPU:
  • PM: 80 threads, no overcommitment, 2.2 GHz+
  • VM: 80 vCPUs, 2.2 GHz+
• RAM: 256 GB
• Disk:
  • 960 GB system disk
  • 1200 GB data disk
• NIC:
  • 1 or 2 GE

Minimum cluster

• 3 PMs or 3 VMs
• CPU:
  • PM: 48 threads, no overcommitment, 2.2 GHz+
  • VM: 48 vCPUs, 2.2 GHz+
• RAM: 128 GB per node
• Disk:
  • 960 GB system disk
  • 1800 GB data disk
• NIC:
  • 2 to 4 GE

Notes:

• Network configuration verification and runbook cannot be installed together at this size.
• If AI-based terminal fingerprint identification is required, the minimum cluster grows to 4 nodes.

Distributed cluster

6-node distributed cluster

• 6 VMs total
• 3 service and middleware nodes:
  • 32 vCPUs
  • 96 GB RAM
  • 960 GB disk
• 3 big data nodes:
  • 32 vCPUs
  • 128 GB RAM
  • 300 GB system disk + 1800 GB data disk
• NIC:
  • 2 to 4 GE

9-node distributed cluster

• 9 VMs total
• 3 service nodes:
  • 20 vCPUs
  • 96 GB RAM
  • 960 GB disk
• 3 middleware nodes:
  • 20 vCPUs
  • 64 GB RAM
  • 960 GB disk
• 3 big data nodes:
  • 20 vCPUs
  • 128 GB RAM
  • 300 GB system disk + 4000 GB data disk
• NIC:
  • 2 to 4 GE

17-node distributed cluster

• 17 VMs total
• 7 service nodes:
  • 20 vCPUs
  • 96 GB RAM
  • 960 GB disk
• 5 middleware nodes:
  • 20 vCPUs
  • 64 GB RAM
  • 960 GB disk
• 5 big data nodes:
  • 20 vCPUs
  • 128 GB RAM
  • 300 GB system disk + 7000 GB data disk recommended
• NIC:
  • 2 to 4 GE

Multi-cluster system

• Global cluster:
  • 3 nodes
  • 16 vCPUs per node
  • 64 GB RAM per node
  • 600 GB disk per node
  • 2 to 4 GE NICs
• Regional clusters:
  • are sized according to the selected minimum-cluster or distributed-cluster model
  • cannot be single-node systems

Additional Design Notes

• On-premises supports PM- and VM-based deployment.
• In VM mode, VMware, FusionCompute, and HCS are supported.
• Huawei servers are required for PM-based deployment.
• For Huawei servers with virtualization, FusionCompute or HCS is recommended.
• For third-party servers prepared by the customer, VMware is recommended.
• Public cloud deployment is currently supported on Huawei Cloud ECS.
• SSD-based and HDD-based servers must not be mixed in the same deployment network.
• In VM environments, the overhead of the virtualization platform must be included in planning.
• Huawei gives a typical example of around 8 vCPUs, 32 GB RAM, and 100 GB disk of additional overhead for virtualization software.
• According to newer Hedex documentation, some deployment and DR functions depend on release, edition, and operating model.
• Therefore, multi-cluster, DR, and public cloud capabilities should always be checked against the exact support matrix of the deployed release.

Authentication Component

• Deployed as a single-node system.
• Huawei explicitly recommends PM-based deployment here.
• Typical minimum requirements:
  • 1 PM or 1 VM
  • 20 threads or 20 vCPUs
  • 64 GB RAM
  • 960 GB disk
  • 1 or 2 GE NICs

Key Takeaways

• Single-node = simplest and smallest variant, but with the lowest resilience.
• Minimum cluster = smallest true production-grade cluster variant.
• Distributed cluster = preferred for larger environments and clear role separation.
• Multi-cluster = central management of multiple regional clusters.
• HA in one DC and geographic DR across multiple DCs are two different layers in the design.
• For geographic DR, Huawei recommends an automatic DR solution with arbitration plus separate primary, secondary, and arbitration sites.

Sources

• 001_Docs/IMasterNCE/[iMaster NCE-Campus Encyclopedia] Installation Resource Planning and Requirements.pdf
• 001_Docs/IMasterNCE/【iMaster NCE-Campus Encyclopedia】Installation Scenarios.pdf
• 001_Docs/IMasterNCE/profile.xml
• 001_Docs/IMasterNCE/resources/

Samuel Heinrich

Senior Network Engineer at Selution AG (Switzerland)

Arbeitet in Raum Basel (Switzerland) als Senior Network Engineer mit über 15 Jahren Erfahrung im Bereich Netzwerk