1.. SPDX-License-Identifier: BSD-3-Clause 2 Copyright(c) 2016 Intel Corporation. 3 4.. _virtio_user_for_container_networking: 5 6Virtio_user for Container Networking 7==================================== 8 9Container becomes more and more popular for strengths, like low overhead, fast 10boot-up time, and easy to deploy, etc. How to use DPDK to accelerate container 11networking becomes a common question for users. There are two use models of 12running DPDK inside containers, as shown in 13:numref:`figure_use_models_for_running_dpdk_in_containers`. 14 15.. _figure_use_models_for_running_dpdk_in_containers: 16 17.. figure:: img/use_models_for_running_dpdk_in_containers.* 18 19 Use models of running DPDK inside container 20 21This page will only cover aggregation model. 22 23Overview 24-------- 25 26The virtual device, virtio-user, with unmodified vhost-user backend, is designed 27for high performance user space container networking or inter-process 28communication (IPC). 29 30The overview of accelerating container networking by virtio-user is shown 31in :numref:`figure_virtio_user_for_container_networking`. 32 33.. _figure_virtio_user_for_container_networking: 34 35.. figure:: img/virtio_user_for_container_networking.* 36 37 Overview of accelerating container networking by virtio-user 38 39Different virtio PCI devices we usually use as a para-virtualization I/O in the 40context of QEMU/VM, the basic idea here is to present a kind of virtual devices, 41which can be attached and initialized by DPDK. The device emulation layer by 42QEMU in VM's context is saved by just registering a new kind of virtual device 43in DPDK's ether layer. And to minimize the change, we reuse already-existing 44virtio PMD code (driver/net/virtio/). 45 46Virtio, in essence, is a shm-based solution to transmit/receive packets. How is 47memory shared? In VM's case, qemu always shares the whole physical layout of VM 48to vhost backend. But it's not feasible for a container, as a process, to share 49all virtual memory regions to backend. So only those virtual memory regions 50(aka, hugepages initialized in DPDK) are sent to backend. It restricts that only 51addresses in these areas can be used to transmit or receive packets. 52 53Sample Usage 54------------ 55 56Here we use Docker as container engine. It also applies to LXC, Rocket with 57some minor changes. 58 59#. Write a Dockerfile like below. 60 61 .. code-block:: console 62 63 cat <<EOT >> Dockerfile 64 FROM ubuntu:latest 65 WORKDIR /usr/src/dpdk 66 COPY . /usr/src/dpdk 67 ENV PATH "$PATH:/usr/src/dpdk/<build_dir>/app/" 68 EOT 69 70#. Build a Docker image. 71 72 .. code-block:: console 73 74 docker build -t dpdk-app-testpmd . 75 76#. Start a testpmd on the host with a vhost-user port. 77 78 .. code-block:: console 79 80 $(testpmd) -l 0-1 -n 4 --socket-mem 1024,1024 \ 81 --vdev 'eth_vhost0,iface=/tmp/sock0' \ 82 --file-prefix=host --no-pci -- -i 83 84#. Start a container instance with a virtio-user port. 85 86 .. code-block:: console 87 88 docker run -i -t -v /tmp/sock0:/var/run/usvhost \ 89 -v /dev/hugepages:/dev/hugepages \ 90 dpdk-app-testpmd testpmd -l 6-7 -n 4 -m 1024 --no-pci \ 91 --vdev=virtio_user0,path=/var/run/usvhost \ 92 --file-prefix=container \ 93 -- -i 94 95Note: If we run all above setup on the host, it's a shm-based IPC. 96 97Limitations 98----------- 99 100We have below limitations in this solution: 101 * Cannot work with --huge-unlink option. As we need to reopen the hugepage 102 file to share with vhost backend. 103 * Cannot work with --no-huge option. Currently, DPDK uses anonymous mapping 104 under this option which cannot be reopened to share with vhost backend. 105 * Cannot work when there are more than VHOST_MEMORY_MAX_NREGIONS(8) hugepages. 106 If you have more regions (especially when 2MB hugepages are used), the option, 107 --single-file-segments, can help to reduce the number of shared files. 108 * Applications should not use file name like HUGEFILE_FMT ("%smap_%d"). That 109 will bring confusion when sharing hugepage files with backend by name. 110 * Root privilege is a must. DPDK resolves physical addresses of hugepages 111 which seems not necessary, and some discussions are going on to remove this 112 restriction. 113