Hyper-threading and NUMA

This section provides information about hyper-threading, pinning, NICs in the NUMA node, and core sibling information inside the DU Worker node.

Hyper-threading

If hyper-threading is enabled, then each core is logically divided into two hyper-threads or physical CPUs (pCPU):

Core	pCPU
Core 0	pCPU 0 and pCPU 1
Core 1	pCPU 2 and pCPU 3
Core 2	pCPU 4 and pCPU 5
And so on.

Creating and Pinning 40 vCPU DU Worker Nodes

The following architectural diagram provides information about how a Worker node vCPU can be pinned to a pCPU if there are no other VMs running on the host.

Note: After you pin a DU worker node, it does not move across the pCPU. You can configure pinning using isNumaConfigNeeded flag in the CSAR file. This flag must be set to true.

NICs in NUMA

When the DU worker node is requesting for I/O devices through CSAR, it can either choose the I/O connected to the same NUMA or share the I/O with different NUMA. This is configured using isSharedAcrossNuma flag in the CSAR file. If this flag set to true, it can source the I/O devices from a different NUMA node. If this flag is set to false or not present, it will source I/O devices connected to the same NUMA to which the DU worker node is pinned.

Core Sibling Information Inside DU Worker Node

To expose hyper-threading details inside the VM, we must enable the vHT feature through CSAR. After you enable vHT, the Worker node VM can access hyper-threading sibling relations inside the VM. After enabling vHT, The lscpu -e -a command displays all vCPUs and its associated core and socket. The lscpu command also displays the threads per core information.

The following command provides sibling information:

cat /sys/devices/system/cpu/cpu'x'/topology/thread_siblings_list

For information about enabling vHT, see Enable Virtual Hyper-Threading.