The High Precision Timer (HPET) must be enabled in the platform BIOS if the HPET is to be used. Otherwise, the Time Stamp Counter (TSC) is used by default. The BIOS is typically accessed by pressing F2 while the platform is starting up. The user can then navigate to the HPET option. On the Crystal Forest platform BIOS, the path is: Advanced -> PCH-IO Configuration -> High Precision Timer -> (Change from Disabled to Enabled if necessary).
On a system that has already booted, the following command can be issued to check if HPET is enabled:
# grep hpet /proc/timer_list
If no entries are returned, HPET must be enabled in the BIOS (as per the instructions above) and the system rebooted.
The DPDK makes use of the platform HPET timer by mapping the timer counter into the process address space, and as such, requires that the HPET_MMAP kernel configuration option be enabled.
Warning
On Fedora*, and other common distributions such as Ubuntu*, the HPET_MMAP kernel option is not enabled by default. To recompile the Linux kernel with this option enabled, please consult the distributions documentation for the relevant instructions.
By default, HPET support is disabled in the DPDK build configuration files. To use HPET, the CONFIG_RTE_LIBEAL_USE_HPET setting should be changed to “y”, which will enable the HPET settings at compile time.
For an application to use the rte_get_hpet_cycles() and rte_get_hpet_hz() API calls, and optionally to make the HPET the default time source for the rte_timer library, the new rte_eal_hpet_init() API call should be called at application initialization. This API call will ensure that the HPET is accessible, returning an error to the application if it is not, for example, if HPET_MMAP is not enabled in the kernel. The application can then determine what action to take, if any, if the HPET is not available at run-time.
Note
For applications that require timing APIs, but not the HPET timer specifically, it is recommended that the rte_get_timer_cycles() and rte_get_timer_hz() API calls be used instead of the HPET-specific APIs. These generic APIs can work with either TSC or HPET time sources, depending on what is requested by an application call to rte_eal_hpet_init(), if any, and on what is available on the system at runtime.
Although applications using the DPDK use network ports and other hardware resources directly, with a number of small permission adjustments it is possible to run these applications as a user other than “root”. To do so, the ownership, or permissions, on the following Linux file system objects should be adjusted to ensure that the Linux user account being used to run the DPDK application has access to them:
Note
On some Linux installations, /dev/hugepages is also a hugepage mount point created by default.
Enhanced Intel SpeedStep® Technology must be enabled in the platform BIOS if the power management feature of DPDK is to be used. Otherwise, the sys file folder /sys/devices/system/cpu/cpu0/cpufreq will not exist, and the CPU frequency- based power management cannot be used. Consult the relevant BIOS documentation to determine how these settings can be accessed.
For example, on some Intel reference platform BIOS variants, the path to Enhanced Intel SpeedStep® Technology is:
Advanced->Processor Configuration->Enhanced Intel SpeedStep® Tech
In addition, C3 and C6 should be enabled as well for power management. The path of C3 and C6 on the same platform BIOS is:
Advanced->Processor Configuration->Processor C3 Advanced->Processor Configuration-> Processor C6
While the threads used by an DPDK application are pinned to logical cores on the system, it is possible for the Linux scheduler to run other tasks on those cores also. To help prevent additional workloads from running on those cores, it is possible to use the isolcpus Linux* kernel parameter to isolate them from the general Linux scheduler.
For example, if DPDK applications are to run on logical cores 2, 4 and 6, the following should be added to the kernel parameter list:
isolcpus=2,4,6
To run the DPDK Kernel NIC Interface (KNI) sample application, an extra kernel module (the kni module) must be loaded into the running kernel. The module is found in the kmod sub-directory of the DPDK target directory. Similar to the loading of the igb_uio module, this module should be loaded using the insmod command as shown below (assuming that the current directory is the DPDK target directory):
#insmod kmod/rte_kni.ko
Note
See the “Kernel NIC Interface Sample Application” chapter in the DPDK Sample Applications User Guide for more details.
To enable Intel® VT-d in a Linux kernel, a number of kernel configuration options must be set. These include:
In addition, to run the DPDK with Intel® VT-d, the iommu=pt kernel parameter must be used when using igb_uio driver. This results in pass-through of the DMAR (DMA Remapping) lookup in the host. Also, if INTEL_IOMMU_DEFAULT_ON is not set in the kernel, the intel_iommu=on kernel parameter must be used too. This ensures that the Intel IOMMU is being initialized as expected.
Please note that while using iommu=pt is compulsory for igb_uio driver, the vfio-pci driver can actually work with both iommu=pt and iommu=on.
As there might be firmware fixes for performance enhancement in latest version of firmware image, the firmware update might be needed for getting high performance. Check with the local Intel’s Network Division application engineers for firmware updates. The base driver to support firmware version of FVL3E will be integrated in the next DPDK release, so currently the validated firmware version is 4.2.6.
PCI configurations of extended_tag and max _read_requ st_size have big impacts on performance of small packets on 40G NIC. Enabling extended_tag and setting max _read_requ st_size to small size such as 128 bytes provide great helps to high performance of small packets.
These can be done in some BIOS implementations.
For other BIOS implementations, PCI configurations can be changed by using command of setpci, or special configurations in DPDK config file of common_linux.
Bits 7:5 at address of 0xA8 of each PCI device is used for setting the max_read_request_size, and bit 8 of 0xA8 of each PCI device is used for enabling/disabling the extended_tag. lspci and setpci can be used to read the values of 0xA8 and then write it back after being changed.
In config file of common_linux, below three configurations can be changed for the same purpose.
CONFIG_RTE_PCI_CONFIG
CONFIG_RTE_PCI_EXTENDED_TAG
CONFIG_RTE_PCI_MAX_READ_REQUEST_SIZE
As i40e PMD supports both 16 and 32 bytes RX descriptor sizes, and 16 bytes size can provide helps to high performance of small packets. Configuration of CONFIG_RTE_LIBRTE_I40E_16BYTE_RX_DESC in config files can be changed to use 16 bytes size RX descriptors.
Due to the hardware design, the interrupt signal inside NIC is needed for per packet descriptor write-back. The minimum interval of interrupts could be set at compile time by CONFIG_RTE_LIBRTE_I40E_ITR_INTERVAL in configuration files. Though there is a default configuration, the interval could be tuned by the users with that configuration item depends on what the user cares about more, performance or per packet latency.