INSTALL.DPDK

   1                    Using Open vSwitch with DPDK
   2                    ============================
   3
   4 Open vSwitch can use Intel(R) DPDK lib to operate entirely in
   5 userspace. This file explains how to install and use Open vSwitch in
   6 such a mode.
   7
   8 The DPDK support of Open vSwitch is considered experimental.
   9 It has not been thoroughly tested.
  10
  11 This version of Open vSwitch should be built manually with "configure"
  12 and "make".
  13
  14 Building and Installing:
  15 ------------------------
  16
  17 Recommended to use DPDK 1.6.
  18
  19 DPDK:
  20 Set dir i.g.:   export DPDK_DIR=/usr/src/dpdk-1.6.0r2
  21 cd $DPDK_DIR
  22 update config/defconfig_x86_64-default-linuxapp-gcc so that dpdk generate single lib file.
  23 CONFIG_RTE_BUILD_COMBINE_LIBS=y
  24
  25 make install T=x86_64-default-linuxapp-gcc
  26 For details refer to  http://dpdk.org/
  27
  28 Linux kernel:
  29 Refer to intel-dpdk-getting-started-guide.pdf for understanding
  30 DPDK kernel requirement.
  31
  32 OVS:
  33 cd $(OVS_DIR)/openvswitch
  34 ./boot.sh
  35 export DPDK_BUILD=/usr/src/dpdk-1.6.0r2/x86_64-default-linuxapp-gcc
  36 ./configure --with-dpdk=$DPDK_BUILD
  37 make
  38
  39 Refer to INSTALL.userspace for general requirements of building
  40 userspace OVS.
  41
  42 Using the DPDK with ovs-vswitchd:
  43 ---------------------------------
  44
  45 Setup system boot:
  46    kernel bootline, add: default_hugepagesz=1GB hugepagesz=1G hugepages=1
  47
  48 First setup DPDK devices:
  49   - insert uio.ko
  50     e.g. modprobe uio
  51   - insert igb_uio.ko
  52     e.g. insmod DPDK/x86_64-default-linuxapp-gcc/kmod/igb_uio.ko
  53   - Bind network device to ibg_uio.
  54     e.g. DPDK/tools/pci_unbind.py --bind=igb_uio eth1
  55     Alternate binding method:
  56      Find target Ethernet devices
  57       lspci -nn|grep Ethernet
  58      Bring Down (e.g. eth2, eth3)
  59       ifconfig eth2 down
  60       ifconfig eth3 down
  61      Look at current devices (e.g ixgbe devices)
  62       ls /sys/bus/pci/drivers/ixgbe/
  63       0000:02:00.0  0000:02:00.1  bind  module  new_id  remove_id  uevent  unbind
  64      Unbind target pci devices from current driver (e.g. 02:00.0 ...)
  65       echo 0000:02:00.0 > /sys/bus/pci/drivers/ixgbe/unbind
  66       echo 0000:02:00.1 > /sys/bus/pci/drivers/ixgbe/unbind
  67      Bind to target driver (e.g. igb_uio)
  68       echo 0000:02:00.0 > /sys/bus/pci/drivers/igb_uio/bind
  69       echo 0000:02:00.1 > /sys/bus/pci/drivers/igb_uio/bind
  70      Check binding for listed devices
  71       ls /sys/bus/pci/drivers/igb_uio
  72       0000:02:00.0  0000:02:00.1  bind  module  new_id  remove_id  uevent  unbind
  73
  74 Prepare system:
  75   - load ovs kernel module
  76     e.g modprobe openvswitch
  77   - mount hugetlbfs
  78     e.g. mount -t hugetlbfs -o pagesize=1G none /mnt/huge/
  79
  80 Ref to http://www.dpdk.org/doc/quick-start for verifying DPDK setup.
  81
  82 Start ovsdb-server as discussed in INSTALL doc:
  83   Summary e.g.:
  84     First time only db creation (or clearing):
  85       mkdir -p /usr/local/etc/openvswitch
  86       mkdir -p /usr/local/var/run/openvswitch
  87       rm /usr/local/etc/openvswitch/conf.db
  88       cd $OVS_DIR
  89       ./ovsdb/ovsdb-tool create /usr/local/etc/openvswitch/conf.db \
  90         ./vswitchd/vswitch.ovsschema
  91     start ovsdb-server
  92       cd $OVS_DIR
  93       ./ovsdb/ovsdb-server --remote=punix:/usr/local/var/run/openvswitch/db.sock \
  94           --remote=db:OpenOpen_vSwitch,manager_options \
  95           --private-key=db:Open_vSwitch,SSL,private_key \
  96           --certificate=dbitch,SSL,certificate \
  97           --bootstrap-ca-cert=db:Open_vSwitch,SSL,ca_cert --pidfile --detach
  98     First time after db creation, initialize:
  99       cd $OVS_DIR
 100       ./utilities/ovs-vsctl --no-wait init
 101
 102 Start vswitchd:
 103 DPDK configuration arguments can be passed to vswitchd via `--dpdk`
 104 argument. dpdk arg -c is ignored by ovs-dpdk, but it is a required parameter
 105 for dpdk initialization.
 106
 107    e.g.
 108    export DB_SOCK=/usr/local/var/run/openvswitch/db.sock
 109    ./vswitchd/ovs-vswitchd --dpdk -c 0x1 -n 4 -- unix:$DB_SOCK  --pidfile --detach
 110
 111 If allocated more than 1 GB huge pages, set amount and use NUMA node 0 memory:
 112
 113    ./vswitchd/ovs-vswitchd --dpdk -c 0x1 -n 4 --socket-mem 1024,0 \
 114       -- unix:$DB_SOCK  --pidfile --detach
 115
 116 To use ovs-vswitchd with DPDK, create a bridge with datapath_type
 117 "netdev" in the configuration database.  For example:
 118
 119     ovs-vsctl add-br br0
 120     ovs-vsctl set bridge br0 datapath_type=netdev
 121
 122 Now you can add dpdk devices. OVS expect DPDK device name start with dpdk
 123 and end with portid. vswitchd should print number of dpdk devices found.
 124
 125     ovs-vsctl add-port br0 dpdk0 -- set Interface dpdk0 type=dpdk
 126     ovs-vsctl add-port br0 dpdk1 -- set Interface dpdk1 type=dpdk
 127
 128 Once first DPDK port is added to vswitchd, it creates a Polling thread and
 129 polls dpdk device in continuous loop. Therefore CPU utilization
 130 for that thread is always 100%.
 131
 132 Test flow script across NICs (assuming ovs in /usr/src/ovs):
 133   Assume 1.1.1.1 on NIC port 1 (dpdk0)
 134   Assume 1.1.1.2 on NIC port 2 (dpdk1)
 135   Execute script:
 136
 137 ############################# Script:
 138
 139 #! /bin/sh
 140
 141 # Move to command directory
 142
 143 cd /usr/src/ovs/utilities/
 144
 145 # Clear current flows
 146 ./ovs-ofctl del-flows br0
 147
 148 # Add flows between port 1 (dpdk0) to port 2 (dpdk1)
 149 ./ovs-ofctl add-flow br0 in_port=1,dl_type=0x800,nw_src=1.1.1.1,\
 150 nw_dst=1.1.1.2,idle_timeout=0,action=output:2
 151 ./ovs-ofctl add-flow br0 in_port=2,dl_type=0x800,nw_src=1.1.1.2,\
 152 nw_dst=1.1.1.1,idle_timeout=0,action=output:1
 153
 154 ######################################
 155
 156 Ideally for maximum throughput, the 100% task should not be scheduled out
 157 which temporarily halts the process. The following affinitization methods will
 158 help.
 159
 160 At this time all ovs-vswitchd tasks end up being affinitized to cpu core 0
 161 but this may change. Lets pick a target core for 100% task to run on, i.e. core 7.
 162 Also assume a dual 8 core sandy bridge system with hyperthreading enabled.
 163 (A different cpu configuration will have different core mask requirements).
 164
 165 To give better ownership of 100%, isolation maybe useful.
 166 To kernel bootline add core isolation list for core 7 and associated hype core 23
 167   e.g.  isolcpus=7,23
 168 Reboot system for isolation to take effect, restart everything
 169
 170 List threads (and their pid) of ovs-vswitchd
 171   top -p `pidof ovs-vswitchd` -H -d1
 172
 173 Look for pmd* thread which is polling dpdk devices, this will be the 100% CPU
 174 bound task. Using this thread pid, affinitize to core 7 (mask 0x080),
 175 example pid 1762
 176
 177 taskset -p 080 1762
 178   pid 1762's current affinity mask: 1
 179   pid 1762's new affinity mask: 80
 180
 181 Assume that all other ovs-vswitchd threads to be on other socket 0 cores.
 182 Affinitize the rest of the ovs-vswitchd thread ids to 0x0FF007F
 183
 184 taskset -p 0x0FF007F {thread pid, e.g 1738}
 185   pid 1738's current affinity mask: 1
 186   pid 1738's new affinity mask: ff007f
 187 . . .
 188
 189 The core 23 is left idle, which allows core 7 to run at full rate.
 190
 191 Future changes may change the need for cpu core affinitization.
 192
 193 Restrictions:
 194 -------------
 195
 196   - This Support is for Physical NIC. I have tested with Intel NIC only.
 197   - vswitchd userspace datapath does affine polling thread but it is
 198     assumed that devices are on numa node 0. Therefore if device is
 199     attached to non zero numa node switching performance would be
 200     suboptimal.
 201   - There are fixed number of polling thread and fixed number of per
 202     device queues configured.
 203   - Work with 1500 MTU, needs few changes in DPDK lib to fix this issue.
 204   - Currently DPDK port does not make use any offload functionality.
 205
 206 Bug Reporting:
 207 --------------
 208
 209 Please report problems to bugs@openvswitch.org.