1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2015 Intel Corporation.
4 How to get best performance with NICs on Intel platforms
5 ========================================================
7 This document is a step-by-step guide for getting high performance from DPDK applications on Intel platforms.
10 Hardware and Memory Requirements
11 --------------------------------
13 For best performance use an Intel Xeon class server system such as Ivy Bridge, Haswell or newer.
15 Ensure that each memory channel has at least one memory DIMM inserted, and that the memory size for each is at least 4GB.
16 **Note**: this has one of the most direct effects on performance.
18 You can check the memory configuration using ``dmidecode`` as follows::
20 dmidecode -t memory | grep Locator
40 The sample output above shows a total of 8 channels, from ``A`` to ``H``, where each channel has 2 DIMMs.
42 You can also use ``dmidecode`` to determine the memory frequency::
44 dmidecode -t memory | grep Speed
47 Configured Clock Speed: 2134 MHz
49 Configured Clock Speed: Unknown
51 Configured Clock Speed: 2134 MHz
55 Configured Clock Speed: 2134 MHz
57 Configured Clock Speed: Unknown
59 Configured Clock Speed: 2134 MHz
61 Configured Clock Speed: Unknown
63 The output shows a speed of 2133 MHz (DDR4) and Unknown (not existing).
64 This aligns with the previous output which showed that each channel has one memory bar.
67 Network Interface Card Requirements
68 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
70 Use a `DPDK supported <https://core.dpdk.org/supported/>`_ high end NIC such as the Intel XL710 40GbE.
72 Make sure each NIC has been flashed the latest version of NVM/firmware.
74 Use PCIe Gen3 slots, such as Gen3 ``x8`` or Gen3 ``x16`` because PCIe Gen2 slots don't provide enough bandwidth
75 for 2 x 10GbE and above.
76 You can use ``lspci`` to check the speed of a PCI slot using something like the following::
78 lspci -s 03:00.1 -vv | grep LnkSta
80 LnkSta: Speed 8GT/s, Width x8, TrErr- Train- SlotClk+ DLActive- ...
81 LnkSta2: Current De-emphasis Level: -6dB, EqualizationComplete+ ...
83 When inserting NICs into PCI slots always check the caption, such as CPU0 or CPU1 to indicate which socket it is connected to.
85 Care should be take with NUMA.
86 If you are using 2 or more ports from different NICs, it is best to ensure that these NICs are on the same CPU socket.
87 An example of how to determine this is shown further below.
93 The following are some recommendations on BIOS settings. Different platforms will have different BIOS naming
94 so the following is mainly for reference:
96 #. Establish the steady state for the system, consider reviewing BIOS settings desired for best performance characteristic e.g. optimize for performance or energy efficiency.
98 #. Match the BIOS settings to the needs of the application you are testing.
100 #. Typically, **Performance** as the CPU Power and Performance policy is a reasonable starting point.
102 #. Consider using Turbo Boost to increase the frequency on cores.
104 #. Disable all virtualization options when you test the physical function of the NIC, and turn on VT-d if you wants to use VFIO.
107 Linux boot command line
108 ~~~~~~~~~~~~~~~~~~~~~~~
110 The following are some recommendations on GRUB boot settings:
112 #. Use the default grub file as a starting point.
114 #. Reserve 1G huge pages via grub configurations. For example to reserve 8 huge pages of 1G size::
116 default_hugepagesz=1G hugepagesz=1G hugepages=8
118 #. Isolate CPU cores which will be used for DPDK. For example::
120 isolcpus=2,3,4,5,6,7,8
122 #. If it wants to use VFIO, use the following additional grub parameters::
124 iommu=pt intel_iommu=on
127 Configurations before running DPDK
128 ----------------------------------
130 1. Reserve huge pages.
131 See the earlier section on :ref:`linux_gsg_hugepages` for more details.
133 .. code-block:: console
135 # Get the hugepage size.
136 awk '/Hugepagesize/ {print $2}' /proc/meminfo
138 # Get the total huge page numbers.
139 awk '/HugePages_Total/ {print $2} ' /proc/meminfo
141 # Unmount the hugepages.
142 umount `awk '/hugetlbfs/ {print $2}' /proc/mounts`
144 # Create the hugepage mount folder.
147 # Mount to the specific folder.
148 mount -t hugetlbfs nodev /mnt/huge
150 2. Check the CPU layout using the DPDK ``cpu_layout`` utility:
152 .. code-block:: console
156 usertools/cpu_layout.py
158 Or run ``lscpu`` to check the cores on each socket.
160 3. Check your NIC id and related socket id:
162 .. code-block:: console
164 # List all the NICs with PCI address and device IDs.
167 For example suppose your output was as follows::
169 82:00.0 Ethernet [0200]: Intel XL710 for 40GbE QSFP+ [8086:1583]
170 82:00.1 Ethernet [0200]: Intel XL710 for 40GbE QSFP+ [8086:1583]
171 85:00.0 Ethernet [0200]: Intel XL710 for 40GbE QSFP+ [8086:1583]
172 85:00.1 Ethernet [0200]: Intel XL710 for 40GbE QSFP+ [8086:1583]
174 Check the PCI device related numa node id:
176 .. code-block:: console
178 cat /sys/bus/pci/devices/0000\:xx\:00.x/numa_node
180 Usually ``0x:00.x`` is on socket 0 and ``8x:00.x`` is on socket 1.
181 **Note**: To get the best performance, ensure that the core and NICs are in the same socket.
182 In the example above ``85:00.0`` is on socket 1 and should be used by cores on socket 1 for the best performance.
184 4. Check which kernel drivers needs to be loaded and whether there is a need to unbind the network ports from their kernel drivers.
185 More details about DPDK setup and Linux kernel requirements see :ref:`linux_gsg_compiling_dpdk` and :ref:`linux_gsg_linux_drivers`.