MeasurementSetup.txt

It is absolutely essential that variability be removed from any performance tests.  This is achieved by controlling for 
known variables in the system and fabric as best as possible.  This is a simple best practices list to aid in config.
========================================================================================================================

Find matching pair of adapters and ports.
 - Check the MAC address to identify the actual port number.
 - Assume the lab wired Port0 to TOR0 and Port1 to TOR1 on all Hosts
 - Using the same port on both physical Hosts avoids TOR0 <-> TOR1 interlink noise and ensures single hop LAT

Check which NUMA node the NIC is under.
 - "Get-NetAdapterRss <NIC_NAME> | fl *", and check for the NUMA value
 - NumaNode                      : 0
 - "Get-NetAdapterRss <NIC_NAME> to view NUMA mapping.
 - "Set-NetAdapterRss <NIC_NAME> -MaxProcessorNumber 9
   - Or similar upper bound for the number of Cores on the associated NUMA node.
   - Note that Cross NUMA traffic is VERY noisy and can result ~50% TPUT difference.

Set system in high-performance mode:
- powercfg.exe /s 8c5e7fda-e8bf-4a96-9a85-a6e23a8c635c

Disable any C-State beyond C1
- powercfg -setacvalueindex scheme_current sub_processor IDLESTATEMAX 1
- powercfg -setactive scheme_current

If using VMs or Hyper-V Containers:
- For Minimum LAT, ensure the VM or Container is on the same NUMA node.
- For Maximum TPUT, ensure the VM or Container is on a separate NUMA node from the Phsical NIC
- For the absolute maximum performance of both TPUT and LAT, ensure the VM/Container are on the 
  same NODE as the Physical NIC, but only if the NUMA node has 20+ Cores.  This is a safe value 
  for 40G perf targets.
- To identify which NUMA node a VM is on, us:
  - Perfmon -> Hper-V VM VID Partition -> Preferred NUMA Node Index 
  - If not NUMA aligned, create a DUMMY VM profile nad start, then start the VM/Container to tests
    and inspect again if it landed on the desired NUMA node.