Para-virtualization

• Paravirtual  RDMA (PVRDMA) is a new  PCIe  virtual NIC

Uses HCA for performance, but works without it
Multiple virtual devices can share an HCA without SR-IOV
Supports  vMotion  (live migration)
https://www.openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf

• Para-virtualization has been commonly used in virtualized environments to improve system efficiency and to optimize management workloads. In the era of Performance Cloud Computing and Big Data Use Cases, cloud providers and data centers focus on developing paravirtualization solutions that provide fast and efficient I/O. Thanks to its nature of high bandwidth, low latency and kernel bypass, Remote Direct Memory Access (RDMA) interconnects are now widely adopted in HPC and Cloud centers as an I/O performance booster. To benefit from these advantages that RDMA offer, network communication supporting InfiniBand and RDMA over Converged Ethernet (RoCE) must be made ready for the underlying virtualized devices

https://www.mikelangelo-project.eu/technology/vrdma-virtio-remote-direct-memory-access/

•   eBPF - extended Berkeley Packet Filter

The Berkeley Packet Filter (BPF) started (article 1992) as a special-purpose virtual machine (register based filter evaluator) for filtering network packets, best known for its use in tcpdump. 

Areas using eBPF:

        XDP - eXpress Data Path

        Traffic control

        Sockets

        Firewalling (xt_bpf module)

        Tracing

        Tracepoints

        kprobe (dynamic tracing of a kernel function call)

        cgroups

https://prototype-kernel.readthedocs.io/en/latest/bpf/

• XDP or eXpress Data Path provides a high performance, programmable network data path in the Linux kernel. XDP provides bare metal packet processing at the lowest point in the software stack. Much of the huge speed gain comes from processing RX packet-pages directly out of drivers RX ring queue before any allocations of meta-data structures like SKBs occur

https://prototype-kernel.readthedocs.io/en/latest/networking/XDP/introduction.html#what-is-xdp

• XDP or eXpress Data Path provides a high performance, programmable network data path in the Linux kernel as part of the IO Visor Project. Use cases for XDP include the following:

    Pre-stack processing like filtering to support DDoS mitigation

    Forwarding and load balancing

    Batching techniques such as in Generic Receive Offload

    Flow sampling, monitoring

    ULP processing (i.e. message delineation)

XDP and DPDK

XDP is sometimes juxtaposed with DPDK when both are perfectly fine approaches. XDP offers another option for users who want performance while still leveraging the programmability of the kernel. Some of the functions that XDP delivers include the following:

    Removes the need for 3rd party code and licensing

    Allows option of busy polling or interrupt driven networking

    Removes the need to allocate large pages

    Removes the need for dedicated CPUs as users have more options on structuring work between CPUs

    Removes the need to inject packets into the kernel from a 3rd party userspace application

    Removes the need to define a new security model for accessing networking hardware

https://www.iovisor.org/technology/xdp

• DPDK is a set of libraries and optimized NIC drivers for fast packet processing in userspace. DPDK provides a framework and common API for high-speed networking applications.Data Plane Development Kit (DPDK)

http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk

• DPDK is a set of libraries and drivers for fast packet processing.

It supports many processor architectures and both FreeBSD and Linux.
https://github.com/spdk/dpdk

• DPDK is a set of libraries and drivers for fast packet processing.It is designed to run on any processors. The first supported CPU was Intel x86 and it is now extended to IBM POWER and ARM.

It runs mostly in Linux userland. A FreeBSD port is available for a subset of DPDK features.
DPDK is not a networking stack and does not provide functions such as Layer-3 forwarding, IPsec, firewalling, etc. Within the tree, however, various application examples are included to help developing such features.
http://www.dpdk.org/

• The DPDK uses the SR-IOV feature for hardware-based I/O sharing in IOV mode. Therefore, it is possible to partition SR-IOV capability on Ethernet controller NIC resources logically and expose them to a virtual machine as a separate PCI function called a “Virtual Function”

https://doc.dpdk.org/guides/nics/intel_vf.html

• SR-IOV Introduce 

The SR-IOV capable device provides a configurable number of independent Virtual Functions, each with its own PCI Configuration space. The Hypervisor assigns one or more Virtual Functions to a virtual machine by mapping the actual configuration space the VFs to the configuration space presented to the virtual machine by the VMM
https://www.slideshare.net/lnxfei/sr-iovlkong

• The Single Root I/O Virtualization (SR-IOV) standard being implemented by 10G network adapter vendors provides hardware support for virtualization,

https://blog.sflow.com/2009/10/sr-iov.html

• Today SR-IOV is the standard framework for PCI Expressdevices

Intel Scalable I/O Virtualization

VF directly assignable to
 Traditional Virtual Machine (VM)
 Bare metal container/process
 VM container

https://www.lfasiallc.com/wp-content/uploads/2017/11/Intel%C2%AE-Scalable-I_O-Virtualization_Kevin-Tian.pdf