DPVS is a high performance Layer-4 load balancer based on DPDK. It's derived from Linux LVS and it's modification alibaba/LVS.

the name DPVS comes from "DPDK-LVS".

Different techniques are applied for high performance:

• Kernel by-pass (user space implementation)
• Share-nothing, per-CPU for key data (Lockless)
• RX Steering and CPU affinity (avoid context switch)
• Batching TX/RX
• Zero Copy (avoid packet copy and syscalls).
• Polling instead of interrupt.
• lockless message for high performance ICP.
• other techs enhanced by DPDK

Major features of DPVS including:

• L4 Load Balancer, including FNAT, DR mode, etc.
• Different schedule algorithm like RR, WLC, WRR, etc.
• User-space Lite IP stack (IPv4, Routing, ARP, ICMP ...).
• SNAT mode for Internet access from internal network.
• Support KNI, VLAN, Bonding for different IDC environment.
• Security aspect, support TCP syn-proxy, Conn-Limit, black-list
• QoS: Traffic Control (Ongoing)

DPVS feature modules are illustrated as following picture.

This quick start is tested with the environment below.

• Linux Distribution: CentOS 7.2
• Kernel: 3.10.0-327.el7.x86_64
• CPU: Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz
• NIC: Intel X540
• Memory: 64G and NUMA system.
• GCC: gcc version 4.8.5 20150623 (Red Hat 4.8.5-4)

Other environment should also OK if DPDK works, pls check dpdk.org for more info.

$ git clone https://github.com/iqiyi/dpvs.git
$ cd dpvs

Well, let's start from DPDK then.

Currently, DPDK-16.07 is used for DPVS and latest DPDK stable version will be integrated soon.

You can skip this section if experienced with DPDK, and refer the link for details.

$ wget https://fast.dpdk.org/rel/dpdk-16.07.2.tar.xz   # download from dpdk.org if link failed.
$ tar vxf dpdk-16.07.2.tar.xz

There's a patch for DPDK kni driver for hardware multicast, apply it if needed (for example, launch ospfd on kni device).

assuming we are in DPVS root dir and dpdk-stable-16.07.2 is under it, pls note it's not mandatory, just for convenience.

$ cd <path-of-dpvs>
$ cp patch/dpdk-16.07/0001-kni-use-netlink-event-for-multicast-driver-part.patch dpdk-stable-16.07.2/
$ cd dpdk-stable-16.07.2/
$ patch -n 1 < 0001-kni-use-netlink-event-for-multicast-driver-part.patch

Now build DPDK and export RTE_SDK env variable for DPDK app (DPVS).

$ cd dpdk-stable-16.07.2/
$ make config T=x86_64-native-linuxapp-gcc
Configuration done
$ make # or make -j40 to save time, where 40 is the cpu core number.
$ export RTE_SDK=$PWD

In our tutorial, RTE_TARGET is not set, the value is "build" by default, thus DPDK libs and header files can be found in dpdk-stable-16.07.2/build. It could be other values.

Now to set up DPDK hugepage, our test environment is NUMA system. For single-node system pls refer the link.

$ # for NUMA machine
$ echo 1024 > /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages
$ echo 1024 > /sys/devices/system/node/node1/hugepages/hugepages-2048kB/nr_hugepages

$ mkdir /mnt/huge
$ mount -t hugetlbfs nodev /mnt/huge

Install Kernel modules and bind NIC with igb_uio driver. Quick start uses only one NIC, normally we use 2 for Full-NAT cluster, even 4 for bonding mode. Assuming eth0 will be used for DPVS/DPDK, and another standalone Linux NIC for debug, for example, eth1.

$ modprob uio
$ cd dpdk-stable-16.07.2/

$ insmod build/kmod/igb_uio.ko
$ insmod build/kmod/rte_kni.ko

$ ./tools/dpdk-devbind.py --st
$ ifconfig eth0 down  # assuming eth0 is 0000:06:00.0
$ ./tools/dpdk-devbind.py -b igb_uio 0000:06:00.0

dpdk-devbind.py -u can be used to unbind driver and switch it back to Linux driver like ixgbe. You can also use lspci or ethtool -i eth0 to check the NIC PCI bus-id. Pls see DPDK site for details.

It's simple, just set RTE_SDK and build it.

$ cd dpdk-stable-16.07.2/
$ export RTE_SDK=$PWD
$ cd <path-of-dpvs>

$ make # or "make -j40" to speed up.
$ make install

may need install dependencies, like openssl and popt.

Output files are installed to dpvs/bin.

$ ls bin/
dpip  dpvs  ipvsadm  keepalived

• dpvs is the main program.
• dpip is the tool to set IP address, route, vlan, neigh etc.
• ipvsadm and keepalived come from LVS, both are modified.

Now, dpvs.conf must be put at /etc/dpvs.conf, just copy it from conf/dpvs.conf.single-nic.sample.

$ cp conf/dpvs.conf.single-nic.sample /etc/dpvs.conf

and start DPVS,

$ cd <path-of-dpvs>/bin
$ ./dpvs &

Check if it's get started ?

$ ./dpip link show
1: dpdk0: socket 0 mtu 1500 rx-queue 8 tx-queue 8
    UP 10000 Mbps full-duplex fixed-nego promisc-off
    addr A0:36:9F:9D:61:F4 OF_RX_IP_CSUM OF_TX_IP_CSUM OF_TX_TCP_CSUM OF_TX_UDP_CSUM

If you see this message. Well done, DPVS is working with NIC dpdk0!

Don't worry if you see this error,

EAL: Error - exiting with code: 1
  Cause: ports in DPDK RTE (2) != ports in dpvs.conf(1)

it means the NIC used by DPVS is not match /etc/dpvs.conf. Pls use dpdk-devbind to adjust the NIC number or modify dpvs.conf. We'll improve this part to make DPVS more "clever" to avoid modify config file when NIC count is not match.

The test topology looks like,

Set VIP and Local IP (LIP, needed by Full-NAT mode) on DPVS. Let's put commands into setup.sh. You do some check by ./ipvsadm -ln, ./dpip addr show.

$ cat setup.sh
VIP=192.168.100.100
LIP=192.168.100.200
RS=192.168.100.2

./dpip addr add ${VIP}/24 dev dpdk0
./ipvsadm -A -t ${VIP}:80 -s rr
./ipvsadm -a -t ${VIP}:80 -r ${RS} -b

./ipvsadm --add-laddr -z ${LIP} -t 192.168.100.100:80 -F dpdk0
$

$ ./setup.sh

Access VIP from Client, it looks good!

client $ curl 192.168.100.100
Your ip:port : 192.168.100.3:56890

Our test shows the forwarding speed (pps) of DPVS is several times than LVS and as good as Google's Maglev.

Pls see the License file.

DPVS is developing by iQiYi QLB team since Arpil 2016 and now open-sourced. It's already widely used in iQiYi IDC for L4 load balancer and SNAT clusters, and we plan to replace all our LVS clusters with DPVS. We are very happy that more people can get involved in this project. Welcome to try, report issues and submit pull requests. And pls feel free to contact us through Github or Email.