Intel® Ethernet 700/800 Series

Windows Performance Tuning Guide

ID Date Version Classification
784543 02/06/2024 1.1 Public
Document Table of Contents

​Optimization for Specific Usage Models

Table: ​Server Usage Optimization Models provides optimization guidance for specific server usage models.

For best performance, disabling the Firewall is recommended when possible. However, it is understandable that disabling the Firewall is not always possible for production environments and data centers.

General guidance for systems that support more than one NUMA node:

  • ​​Set the preferred NUMA node on each adapter to achieve better scaling across NUMA nodes. Note:​If the preferred NUMA node is not set, the scheduler grabs from any and all available cores, including cross-node, which can have very unreliable results.
  • ​​Pin the application threads to the CPU cores on the NUMA node local to adapter.

    See ​CPU Affinity.

  • Set the RSS Base Processor to a CPU core on the local NUMA node.

    See ​RSS Base Processor Number.

It is generally best to avoid Core 0 due to the administrative overhead tasks processed by the OS on Core 0.

Note:​Some settings are applied system-wide and can impact performance across all Ethernet devices on a server. Settings such as BIOS, Power Management, and Firewall are impacted.

​Server Usage Optimization Models

Optimize For Useful For Optimization Tasks
Quick response and Low Latency Video, audio, and High Performance Computing Cluster (HPCC) servers
  • ​​Disable hyper-threading.
  • ​​Minimize or disable interrupt moderation rate.
  • ​​Disable offload TCP segmentation.
  • ​​Disable jumbo packets.
  • ​​Increase transmit descriptors.
  • ​​Increase receive descriptors.
  • ​​Increase RSS queues.
  • ​​Use an established benchmark like:

    netperf -t TCP_​RR

    netperf -t UDP_​RR

  • ​​Pin benchmark to a single core on NUMA node local to the adapter.
Throughput Data backup/retrieval and file servers
  • ​​Enable jumbo packets.1
  • ​​Increase transmit descriptors.2
  • ​​Increase receive descriptors. 2
  • ​​On systems that support NUMA, set the Preferred NUMA Node on each adapter to achieve better scaling across NUMA nodes.
CPU utilization Application, web, mail, and database servers
  • ​​Maximize interrupt moderation rate.
  • ​​Keep the default setting for the number of receive descriptors; avoid setting large numbers of receive descriptors.
  • ​​Decrease RSS queues.
  • ​​In Hyper-V environments, decrease the maximum number of RSS CPUs.
IP Forwarding Single port to port forwarding
  • ​​Disable hyper-threading.
  • ​​Disable hardware-assisted virtualization, such as Intel® VT-d (or IOMMU) and Intel® AMD-V.
  • ​​Turn off power management.
  • ​​Minimize or disable interrupt moderation rate.
  • ​​Increase transmit descriptors.
  • ​​Increase receive descriptors.
  • ​​Enable port forwarding
Virtualization SR-IOV, IOMMU, Hyper-V, VMs
  • ​​Can achieve efficient resource utilization, improved performance, and enhanced security.
  • ​​Enable Virtualization in BIOS.
  • ​​Enable hardware-assisted virtualization, such as Intel® VT-d (or IOMMU) and Intel® AMD-V.
  • ​​Enable SR-IOV on the interface in device manager, Hyper-V vSwitch manager, and VM hardware acceleration settings.3
  • ​​Allocate appropriate hardware resource.
  • ​​Use dynamic memory allocation for VMs.
  • ​​Proper storage configuration.
  • ​​Optimize networking.
Teaming Configure 2x 700 or 800 series interfaces using a supported Teaming method
  • ​​Teaming ensures uninterrupted connectivity, provides load balancing, and improves overall network reliability.
  • ​​Teaming can be configured using GUI, PowerShell, or Intel® ANS.
  • ​​Update supported drivers and NVM/firmware.
  • ​​Adjust team member properties.
  • ​​Select the appropriate teaming mode.
  • ​​Configure load balancing algorithms.
Notes:
  1. Enable jumbo frames for workloads that require large amounts of data that need to be transferred efficiently. See ​BIOS Settings.
  2. In situations where low RX, low TX, or dropped packets are observed, experiment with increasing the RX and TX buffer sizes. See ​Low Receive (RX) Performance and/or Discarded Packets.
  3. Best performance is generally achieved with SR-IOV enabled, using VF(s) for the SR-IOV enabled network adapter. For environments with SR-IOV disabled, VMQ can be used with standard virtual network adapters to enhance network performance for VMs. See ​SR-IOV/VMs.