About Configuring Lustre Connectivity

Lustre file system is deployed in a subnet, in a customer managed VCN. Lustre clients mount the file system via VCN networking. The networking aspects such as peering, IP addresses, security groups, security lists, and routing are managed by you the customer. See OCI Networking Overview in the Explore More section to learn more about OCI Networking components.

Lustre components are deployed in a customer managed subnet (this subnet must exist during Lustre file system creation). This subnet should have sufficient number of available IP addresses to assign to file system resources, and enough for future growth. The requirement is between 43 to 60 IP addresses per 100TB depending on the performance tier selected. Future file system expansion should be considered while sizing the subnet requirements.

Lustre file system can have its own subnet, or its own VCN depending on size, performance tier, and future expansion requirements.

Lustre can connect to your workloads (Lustre clients) in multiple ways. It is recommended for Lustre to have its own subnet, either in the same VCN as the workload, or a separate VCN. This provides the flexibility to allocate IPs, allocate IP addresses to Lustre workloads separately, and to scale subnet IP space independently.

The different subnet choices do not add additional latency between clients and the file system, as long as the subnet is in the same availability domain. Here are some deployment options for Lustre:

  • Same subnet for Lustre and the workload
  • Lustre and workload in different subnets in the same VCN
  • Lustre and workload in different subnets in different VCNs

If Lustre and workloads are in different VCNs, here are some connectivity options:

  • VCN peering between VCNs in the same tenancy, or across tenancies
  • Create a second VNIC on the workload instance and attach the VNIC to the Lustre subnet, or to a subnet in the Lustre VCN

The following diagram illustrates a deployment using one VCN, with different workload and Lustre subnets:



The following diagram illustrates a Lustre subnet with no workload subnet.



The following diagram illustrates different VCNs for the workload subnet and Lustre subnet.



Here's an illustration of Lustre and workloads in different VCNs, where you can implement VCN peering between VCNs in the same tenancy or across tenancies. A second VNIC is created on the workload instance and attached to the Lustre subnet, or to a subnet in the Lustre VCN.



Considerations for Availability Domains and Network Latency

Lustre file system can be mounted from anywhere, as long as there is network connectivity. This includes from other regions and on-premises. However, accessing the file system across high latency links will encounter performance issues. The extent of the performance degradation depends on the link bandwidth and network latency between Lustre and clients.