Lingjun network includes basic widgets such as Lingjun CIDR blocks, Lingjun connection instances, and Lingjun hubs. You can create Lingjun CIDR blocks and Lingjun subnets based on your business requirements. Deploy Lingjun nodes in Lingjun subnets to connect them to Cloud Enterprise Network (CEN) or your VPCs by using Lingjun connection instances. You can use Lingjun hubs to connect different Lingjun CIDR blocks and connect Lingjun CIDR blocks to Lingjun connection instances.
Note: This document applies only to node instance types that require configuring Lingjun connection instances when creating a cluster.
Network architecture
It contains the following widgets:
Lingjun CIDR block: the private network space of the Lingjun network, used for planning the overall IP address range.
Lingjun subnets: a subnet in the Lingjun CIDR block, used to deploy and connect Lingjun nodes.
Lingjun connection instances: connects Lingjun nodes to Cloud Enterprise Network (CEN) and VPC networks.
Lingjun HUB: connects different Lingjun CIDR blocks, or connects Lingjun CIDR blocks to Lingjun connection instances.
Lingjun network interface controllers (NICs): A network interface on a Lingjun node for inter-node communication.
CEN: Enables private network intercommunication between VPCs and between VPCs and on-premises data centers.
How many Lingjun CIDR blocks should you use?
Scenario 1: RDMA communication required
If RDMA communication is required between nodes, these nodes must be deployed in the same Lingjun CIDR block. RDMA does not support cross-Lingjun CIDR block communication.
Scenario 2: RDMA communication not required
If RDMA is not involved, determine whether network isolation is required:
No isolation requirement: We recommend using only 1 Lingjun CIDR block for simpler management.
Isolation required: For example, to separate the production environment from the test environment, we recommend using multiple Lingjun CIDR blocks.
When using multiple Lingjun CIDR blocks, note that:
Multiple clusters can be deployed within one Lingjun CIDR block.
A Lingjun cluster can only be deployed in one Lingjun CIDR block.
Nodes in different Lingjun CIDR blocks cannot form the same cluster through RDMA.
Plan CIDR blocks in advance to avoid address conflicts.
To enable intercommunication between different CIDR blocks, configure a routing policy through Lingjun HUB.
How many Lingjun subnets should you use?
Plan Lingjun subnets based on the number of network interface cards (NICs). It is recommended to connect multiple NICs on the same Lingjun node to different subnets. If too many Lingjun NICs are connected to a single subnet, there may be too many ARP entries on the node, which can affect RDMA communication stability and, in critical cases, cause training job interruptions.
Example: If a Lingjun node has 5 Lingjun NICs, it is recommended to create 5 Lingjun subnets under the corresponding Lingjun CIDR block, with each NIC connected to a different subnet.
The communication latency and throughput between different Lingjun subnets within the same zone in the same Lingjun CIDR block are the same as those within the same subnet.
How do I specify CIDR blocks?
When you create Lingjun VPDs and Lingjun subnets, you must specify a private CIDR block for your Lingjun network in CIDR block format.
Plan Lingjun VPDs
You can use any CIDR block and its subnets as the CIDR block of a Lingjun VPD, except 100.64.0.0/10, 224.0.0.0/4, 127.0.0.0/8, and 169.254.0.0/16.
You must plan a Lingjun VPD in advance. The Lingjun VPD cannot conflict with the CIDR blocks of other networks to which the Lingjun cluster is to be connected, such as CIDR blocks of VPCs or data centers.
The number of available IP addresses in a Lingjun VPD determines the maximum number of nodes that can be deployed in the Lingjun cluster. You must prepare a CIDR block whose subnet mask length is greater than 22 bits to prevent the cluster from being unable to be scaled up.
Plan Lingjun Subnets
The CIDR block of a Lingjun subnet must be a subset of the Lingjun CIDR block to which it belongs. For example, if the Lingjun CIDR block is 192.168.0.0/16, the CIDR block of a Lingjun subnet under that Lingjun CIDR block can range from 192.168.0.0/17 to 192.168.0.0/29.
The mask length of a Lingjun subnet is recommended to be no less than 23 (meaning a single Lingjun subnet contains no more than 512 IP addresses), to avoid excessive ARP entries for address resolution on Lingjun nodes, which could lead to resource exhaustion on Lingjun nodes.
When planning the CIDR blocks of Lingjun subnets, also consider the number of IP addresses required for Lingjun nodes in the subnet.