VMware Performance Tuning

The following sections contain VMware performance tuning recommendations to improve system performance. These performance recommendations are general guidelines and are not all-inclusive.

BIOS Setting Recommendations

Ribbon recommends the following BIOS settings for optimum performance. The following table is intended as a reference. Exact values may vary based on vendor and HW.

The BIOS settings shown in the example below are recommended for HP DL380p Gen8 servers. For BIOS settings of other servers, refer to the respective vendor's website.

General Recommendations

• Ensure the number of vCPUs in an instance is always an even number (4, 6, 8, and so on), as hyper threaded vCPUs are used.
• For best performance, make sure a single instance is confined to a single NUMA. Performance degradation occurs if an instance spans across multiple NUMAs.
• Ensure the physical NICs associated with an instance are connected to the same NUMA/socket where the instance is hosted. Doing so reduces the remote node memory access, which in turn helps improve the performance.

  • Log in to the ESX host

  • Check the NIC in use by using the esxcli network nic list command.

  • Find out the NUMA affinity of the NIC using the command vsish -e get /net/pNics/<vmnicx>/properties | grep "NUMA".

ESXi Host Configuration Parameters

Use the VMware vSphere client to configure the following ESXi host configuration parameters on the Advanced Settings page (see figure below) before installing the SBC SWe.

Path: Host > Manage > Advanced Settings

VM Settings

CPU Settings

To edit CPU, go to Edit instance settings > Virtual Hardware > CPU.

Recommended CPU Settings

Memory Settings

To edit Memory, go to Edit instance settings > Virtual Hardware > Memory.

Recommended Memory Settings

Latency Sensitivity Settings

To edit Latency sensitivity, go to Edit instance settings > VM Options > Advanced > Latency sensitivity.

Configure the VM Latency Sensitivity to High, if the ESX allows it.

NUMA Settings

To edit NUMA settings, go to Edit instance settings > VM Options > Advanced > Configuration Parameters > Edit Configuration.

Configure numa.nodeAffinity based on the NUMA node to which Pkt NICs are attached (as mentioned in General Recommendations). Ensure the VM memory fits in a single NUMA node, so that remote memory access does not happen.

Configure numa.vcpu.preferHT=TRUE. This is required for better cache optimizations. Refer to http://www.staroceans.org/ESXi_VMkernel_NUMA_Constructs.htm for further details.

Configure numa.autosize.once = FALSE.

Setting the Scheduling Affinity of a VMware Instance

By specifying a CPU affinity setting for each virtual machine, you can restrict the assignment of virtual machines to a subset of the available cores. By using this feature, you can assign each virtual machine to cores in the specified affinity set.

Each virtual machine on the Host instance (not just the SBC instance) must have a dedicated "CPU affinity" set, so that one VM will not interfere with another VM.

Procedure

1. Browse to the virtual machine in the vSphere Client.
2. Select VM, right-click on the virtual machine, and click Edit Settings.
3. Under Virtual Hardware, expand CPU.
4. Under Scheduling Affinity, specify the core numbers
   Use '-' for ranges and ',' to separate values.
   For example, "0, 2, 4-7" would indicate CPU cores 0, 2, 4, 5, 6 and 7.
5. Select the cores where you want the virtual machine to run and click Save.

Both logical cores of a physical core need assignment in the "Scheduling affinity" set of the same virtual machine. Logical cores on the same physical core have consecutive CPU numbers, so that CPUs 0 and 1 are on the first core together, CPUs 2 and 3 are on the second core, and so on.

You can let the Host OS run some of the CPU cores.

Validation

Use the below ESXI CLI command to validate the "Scheduling affinity" of the VM:

sched-stats -t cpu| grep vcpu

Example output from the above command:

In the preceding example:

VMware does not provide any method for the user to pin a vCPU thread to a particular logical core. You can only provide a "Scheduling affinity" set for the VM, and the vCPU thread will get scheduled within the given CPU affinity set.