The DPDK framework creates a set of libraries for specific hardware/software environments through the creation of an Environment Abstraction Layer. The EAL hides the environment specifics and provides a standard programming interface to libraries, available hardware accelerators and other hardware and operating system elements. Once the EAL is created for a specific environment, developers link to the library to create their applications. For instance, EAL provides the frameworks to support Linux, FreeBSD, Intel IA-32 or 64-bit, IBM POWER9 and ARM 32- or 64-bit. The EAL also provides additional services including time references, generic bus access, trace and debug functions and alarm operations. The DPDK implements a low overhead run-to-completion model for fast data plane performance and accesses devices via polling to eliminate the performance overhead of interrupt processing. The DPDK is also in the process of including the event-based programming model for fast data plane processing. The DPDK also includes software examples that highlight best practices for software architecture, tips for data structure design and storage, application profiling and performance tuning utilities and tips that address common network performance deficits.
Libraries
The DPDK includes data plane libraries and optimized network interface controller drivers for the following:
A queue manager implements lockless queues
A buffer manager pre-allocates fixed size buffers
A memory manager allocates pools of objects in memory and uses a ring to store free objects; ensures that objects are spread equally on all DRAM channels
Poll mode drivers are designed to work without asynchronous notifications, reducing overhead
A packet framework – a set of libraries that are helpers to develop packet processing
All libraries are stored in the dpdk/lib/librte_* directories
Plugins
The DPDK includes drivers for many hardware types. There have been some additional out-of-tree plugin drivers in the past, which are now considered deprecated.
librte_pmd_vmxnet3.so provides PMD Ethernet layer supporting Vmxnet3 paravirtualized NIC; superseded by full VMXNET3 support in native DPDK.
librte_pmd_memnic_copy.so provides a Virtual PMD Ethernet layer through shared memory based on 2 memory copies of packets
Environment
The DPDK was originally designed to run using a bare-metal mode which is currently deprecated. Actually, DPDK's EAL provides support for Linux or FreeBSD userland application. EAL can be extended in order to support any processors.
Ecosystem
DPDK is now an open-source project under the Linux Foundation, supported by many companies. DPDK is governed by a Governing Board. The technical activities are overlooked by a Technical Board. Beside Intel, which is a contributor to the DPDK, several other vendors also support the DPDK within their products and some offer additional training, support and professional services. The list of vendors who have announced DPDK support includes: 6WIND, ALTEN Calsoft Labs, Advantech, Brocade, Big Switch Networks, Radisys, Tieto, Wind River, and Lanner.
Projects
The pfSense project published a road map on 25 February 2015, in which developer Jim Thompson announced the rewriting of the pfSense core—including pf, network packet forwarding and shaping, link bonding, IPsec—using DPDK: "We have a goal of being able to forward, with packet filtering at rates of at least 14.88Mpps. This is 'line rate' on a 10Gbps interface. There is simply no way to use today's FreeBSD in-kernel stacks for this type of load." Open vSwitch has a limited set of features running userland that can be leveraged to bypass the Linux kernel OVS processing. This use case of OVS with DPDK userland is usually named OVS-DPDK. It is mostly deployed with OpenStack Neutron but it assumes that many features and software-defined networking capabilities of Openstack are disabled. For instance, when OVS-DPDK is used, Neutron provides a lower level of security than when OVS kernel is used. The FD.IO VPP platform is an extensible framework that provides out-of-the-box production quality switch/router functionality. It is the open source version of Cisco's Vector Packet Processing technology: a high performance, packet-processing stack that can run on commodity CPUs, and can leverage the Poll Mode Drivers for both NICs and cryptographic acceleration hardware and libraries. TRex is an open source traffic generator using DPDK. It generates L4-7 traffic based on pre-processing and smart replay of real traffic templates. TRex amplifies both client and server side traffic and can scale to 200 Gbit/s with one UCS using Intel XL710. TRex also supports multiple streams, ability to change any packet field and provides per stream statistics, latency and jitter. DTS is a Python-based framework for functional tests and benchmarks. It is an open-source project, started in 2014, and is hosted on dpdk.org. It supports both software traffic generators like Scapy and dpdk-pktgen, and a hardware traffic generator like Ixia. DPDK has support for several SRIOV network drivers, enabling creating a PF and VFs, and also to launch VMs and assign VFs to them using PCI Passthrough DDP is one of the new advanced features implemented with DPDK. It allows you to load firmware for a device dynamically, without resetting the host.
