423 lines · plain
1.. SPDX-License-Identifier: GPL-2.02 3====================================4HOWTO for the linux packet generator5====================================6 7Enable CONFIG_NET_PKTGEN to compile and build pktgen either in-kernel8or as a module. A module is preferred; modprobe pktgen if needed. Once9running, pktgen creates a thread for each CPU with affinity to that CPU.10Monitoring and controlling is done via /proc. It is easiest to select a11suitable sample script and configure that.12 13On a dual CPU::14 15 ps aux | grep pkt16 root 129 0.3 0.0 0 0 ? SW 2003 523:20 [kpktgend_0]17 root 130 0.3 0.0 0 0 ? SW 2003 509:50 [kpktgend_1]18 19 20For monitoring and control pktgen creates::21 22 /proc/net/pktgen/pgctrl23 /proc/net/pktgen/kpktgend_X24 /proc/net/pktgen/ethX25 26 27Tuning NIC for max performance28==============================29 30The default NIC settings are (likely) not tuned for pktgen's artificial31overload type of benchmarking, as this could hurt the normal use-case.32 33Specifically increasing the TX ring buffer in the NIC::34 35 # ethtool -G ethX tx 102436 37A larger TX ring can improve pktgen's performance, while it can hurt38in the general case, 1) because the TX ring buffer might get larger39than the CPU's L1/L2 cache, 2) because it allows more queueing in the40NIC HW layer (which is bad for bufferbloat).41 42One should hesitate to conclude that packets/descriptors in the HW43TX ring cause delay. Drivers usually delay cleaning up the44ring-buffers for various performance reasons, and packets stalling45the TX ring might just be waiting for cleanup.46 47This cleanup issue is specifically the case for the driver ixgbe48(Intel 82599 chip). This driver (ixgbe) combines TX+RX ring cleanups,49and the cleanup interval is affected by the ethtool --coalesce setting50of parameter "rx-usecs".51 52For ixgbe use e.g. "30" resulting in approx 33K interrupts/sec (1/30*10^6)::53 54 # ethtool -C ethX rx-usecs 3055 56 57Kernel threads58==============59Pktgen creates a thread for each CPU with affinity to that CPU.60Which is controlled through procfile /proc/net/pktgen/kpktgend_X.61 62Example: /proc/net/pktgen/kpktgend_0::63 64 Running:65 Stopped: eth4@066 Result: OK: add_device=eth4@067 68Most important are the devices assigned to the thread.69 70The two basic thread commands are:71 72 * add_device DEVICE@NAME -- adds a single device73 * rem_device_all -- remove all associated devices74 75When adding a device to a thread, a corresponding procfile is created76which is used for configuring this device. Thus, device names need to77be unique.78 79To support adding the same device to multiple threads, which is useful80with multi queue NICs, the device naming scheme is extended with "@":81device@something82 83The part after "@" can be anything, but it is custom to use the thread84number.85 86Viewing devices87===============88 89The Params section holds configured information. The Current section90holds running statistics. The Result is printed after a run or after91interruption. Example::92 93 /proc/net/pktgen/eth4@094 95 Params: count 100000 min_pkt_size: 60 max_pkt_size: 6096 frags: 0 delay: 0 clone_skb: 64 ifname: eth4@097 flows: 0 flowlen: 098 queue_map_min: 0 queue_map_max: 099 dst_min: 192.168.81.2 dst_max:100 src_min: src_max:101 src_mac: 90:e2:ba:0a:56:b4 dst_mac: 00:1b:21:3c:9d:f8102 udp_src_min: 9 udp_src_max: 109 udp_dst_min: 9 udp_dst_max: 9103 src_mac_count: 0 dst_mac_count: 0104 Flags: UDPSRC_RND NO_TIMESTAMP QUEUE_MAP_CPU105 Current:106 pkts-sofar: 100000 errors: 0107 started: 623913381008us stopped: 623913396439us idle: 25us108 seq_num: 100001 cur_dst_mac_offset: 0 cur_src_mac_offset: 0109 cur_saddr: 192.168.8.3 cur_daddr: 192.168.81.2110 cur_udp_dst: 9 cur_udp_src: 42111 cur_queue_map: 0112 flows: 0113 Result: OK: 15430(c15405+d25) usec, 100000 (60byte,0frags)114 6480562pps 3110Mb/sec (3110669760bps) errors: 0115 116 117Configuring devices118===================119This is done via the /proc interface, and most easily done via pgset120as defined in the sample scripts.121You need to specify PGDEV environment variable to use functions from sample122scripts, i.e.::123 124 export PGDEV=/proc/net/pktgen/eth4@0125 source samples/pktgen/functions.sh126 127Examples::128 129 pg_ctrl start starts injection.130 pg_ctrl stop aborts injection. Also, ^C aborts generator.131 132 pgset "clone_skb 1" sets the number of copies of the same packet133 pgset "clone_skb 0" use single SKB for all transmits134 pgset "burst 8" uses xmit_more API to queue 8 copies of the same135 packet and update HW tx queue tail pointer once.136 "burst 1" is the default137 pgset "pkt_size 9014" sets packet size to 9014138 pgset "frags 5" packet will consist of 5 fragments139 pgset "count 200000" sets number of packets to send, set to zero140 for continuous sends until explicitly stopped.141 142 pgset "delay 5000" adds delay to hard_start_xmit(). nanoseconds143 144 pgset "dst 10.0.0.1" sets IP destination address145 (BEWARE! This generator is very aggressive!)146 147 pgset "dst_min 10.0.0.1" Same as dst148 pgset "dst_max 10.0.0.254" Set the maximum destination IP.149 pgset "src_min 10.0.0.1" Set the minimum (or only) source IP.150 pgset "src_max 10.0.0.254" Set the maximum source IP.151 pgset "dst6 fec0::1" IPV6 destination address152 pgset "src6 fec0::2" IPV6 source address153 pgset "dstmac 00:00:00:00:00:00" sets MAC destination address154 pgset "srcmac 00:00:00:00:00:00" sets MAC source address155 156 pgset "queue_map_min 0" Sets the min value of tx queue interval157 pgset "queue_map_max 7" Sets the max value of tx queue interval, for multiqueue devices158 To select queue 1 of a given device,159 use queue_map_min=1 and queue_map_max=1160 161 pgset "src_mac_count 1" Sets the number of MACs we'll range through.162 The 'minimum' MAC is what you set with srcmac.163 164 pgset "dst_mac_count 1" Sets the number of MACs we'll range through.165 The 'minimum' MAC is what you set with dstmac.166 167 pgset "flag [name]" Set a flag to determine behaviour. Current flags168 are: IPSRC_RND # IP source is random (between min/max)169 IPDST_RND # IP destination is random170 UDPSRC_RND, UDPDST_RND,171 MACSRC_RND, MACDST_RND172 TXSIZE_RND, IPV6,173 MPLS_RND, VID_RND, SVID_RND174 FLOW_SEQ,175 QUEUE_MAP_RND # queue map random176 QUEUE_MAP_CPU # queue map mirrors smp_processor_id()177 UDPCSUM,178 IPSEC # IPsec encapsulation (needs CONFIG_XFRM)179 NODE_ALLOC # node specific memory allocation180 NO_TIMESTAMP # disable timestamping181 SHARED # enable shared SKB182 pgset 'flag ![name]' Clear a flag to determine behaviour.183 Note that you might need to use single quote in184 interactive mode, so that your shell wouldn't expand185 the specified flag as a history command.186 187 pgset "spi [SPI_VALUE]" Set specific SA used to transform packet.188 189 pgset "udp_src_min 9" set UDP source port min, If < udp_src_max, then190 cycle through the port range.191 192 pgset "udp_src_max 9" set UDP source port max.193 pgset "udp_dst_min 9" set UDP destination port min, If < udp_dst_max, then194 cycle through the port range.195 pgset "udp_dst_max 9" set UDP destination port max.196 197 pgset "mpls 0001000a,0002000a,0000000a" set MPLS labels (in this example198 outer label=16,middle label=32,199 inner label=0 (IPv4 NULL)) Note that200 there must be no spaces between the201 arguments. Leading zeros are required.202 Do not set the bottom of stack bit,203 that's done automatically. If you do204 set the bottom of stack bit, that205 indicates that you want to randomly206 generate that address and the flag207 MPLS_RND will be turned on. You208 can have any mix of random and fixed209 labels in the label stack.210 211 pgset "mpls 0" turn off mpls (or any invalid argument works too!)212 213 pgset "vlan_id 77" set VLAN ID 0-4095214 pgset "vlan_p 3" set priority bit 0-7 (default 0)215 pgset "vlan_cfi 0" set canonical format identifier 0-1 (default 0)216 217 pgset "svlan_id 22" set SVLAN ID 0-4095218 pgset "svlan_p 3" set priority bit 0-7 (default 0)219 pgset "svlan_cfi 0" set canonical format identifier 0-1 (default 0)220 221 pgset "vlan_id 9999" > 4095 remove vlan and svlan tags222 pgset "svlan 9999" > 4095 remove svlan tag223 224 225 pgset "tos XX" set former IPv4 TOS field (e.g. "tos 28" for AF11 no ECN, default 00)226 pgset "traffic_class XX" set former IPv6 TRAFFIC CLASS (e.g. "traffic_class B8" for EF no ECN, default 00)227 228 pgset "rate 300M" set rate to 300 Mb/s229 pgset "ratep 1000000" set rate to 1Mpps230 231 pgset "xmit_mode netif_receive" RX inject into stack netif_receive_skb()232 Works with "burst" but not with "clone_skb".233 Default xmit_mode is "start_xmit".234 235Sample scripts236==============237 238A collection of tutorial scripts and helpers for pktgen is in the239samples/pktgen directory. The helper parameters.sh file support easy240and consistent parameter parsing across the sample scripts.241 242Usage example and help::243 244 ./pktgen_sample01_simple.sh -i eth4 -m 00:1B:21:3C:9D:F8 -d 192.168.8.2245 246Usage:::247 248 ./pktgen_sample01_simple.sh [-vx] -i ethX249 250 -i : ($DEV) output interface/device (required)251 -s : ($PKT_SIZE) packet size252 -d : ($DEST_IP) destination IP. CIDR (e.g. 198.18.0.0/15) is also allowed253 -m : ($DST_MAC) destination MAC-addr254 -p : ($DST_PORT) destination PORT range (e.g. 433-444) is also allowed255 -t : ($THREADS) threads to start256 -f : ($F_THREAD) index of first thread (zero indexed CPU number)257 -c : ($SKB_CLONE) SKB clones send before alloc new SKB258 -n : ($COUNT) num messages to send per thread, 0 means indefinitely259 -b : ($BURST) HW level bursting of SKBs260 -v : ($VERBOSE) verbose261 -x : ($DEBUG) debug262 -6 : ($IP6) IPv6263 -w : ($DELAY) Tx Delay value (ns)264 -a : ($APPEND) Script will not reset generator's state, but will append its config265 266The global variables being set are also listed. E.g. the required267interface/device parameter "-i" sets variable $DEV. Copy the268pktgen_sampleXX scripts and modify them to fit your own needs.269 270 271Interrupt affinity272===================273Note that when adding devices to a specific CPU it is a good idea to274also assign /proc/irq/XX/smp_affinity so that the TX interrupts are bound275to the same CPU. This reduces cache bouncing when freeing skbs.276 277Plus using the device flag QUEUE_MAP_CPU, which maps the SKBs TX queue278to the running threads CPU (directly from smp_processor_id()).279 280Enable IPsec281============282Default IPsec transformation with ESP encapsulation plus transport mode283can be enabled by simply setting::284 285 pgset "flag IPSEC"286 pgset "flows 1"287 288To avoid breaking existing testbed scripts for using AH type and tunnel mode,289you can use "pgset spi SPI_VALUE" to specify which transformation mode290to employ.291 292Disable shared SKB293==================294By default, SKBs sent by pktgen are shared (user count > 1).295To test with non-shared SKBs, remove the "SHARED" flag by simply setting::296 297 pg_set "flag !SHARED"298 299However, if the "clone_skb" or "burst" parameters are configured, the skb300still needs to be held by pktgen for further access. Hence the skb must be301shared.302 303Current commands and configuration options304==========================================305 306**Pgcontrol commands**::307 308 start309 stop310 reset311 312**Thread commands**::313 314 add_device315 rem_device_all316 317 318**Device commands**::319 320 count321 clone_skb322 burst323 debug324 325 frags326 delay327 328 src_mac_count329 dst_mac_count330 331 pkt_size332 min_pkt_size333 max_pkt_size334 335 queue_map_min336 queue_map_max337 skb_priority338 339 tos (ipv4)340 traffic_class (ipv6)341 342 mpls343 344 udp_src_min345 udp_src_max346 347 udp_dst_min348 udp_dst_max349 350 node351 352 flag353 IPSRC_RND354 IPDST_RND355 UDPSRC_RND356 UDPDST_RND357 MACSRC_RND358 MACDST_RND359 TXSIZE_RND360 IPV6361 MPLS_RND362 VID_RND363 SVID_RND364 FLOW_SEQ365 QUEUE_MAP_RND366 QUEUE_MAP_CPU367 UDPCSUM368 IPSEC369 NODE_ALLOC370 NO_TIMESTAMP371 SHARED372 373 spi (ipsec)374 375 dst_min376 dst_max377 378 src_min379 src_max380 381 dst_mac382 src_mac383 384 clear_counters385 386 src6387 dst6388 dst6_max389 dst6_min390 391 flows392 flowlen393 394 rate395 ratep396 397 xmit_mode <start_xmit|netif_receive>398 399 vlan_cfi400 vlan_id401 vlan_p402 403 svlan_cfi404 svlan_id405 svlan_p406 407 408References:409 410- ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/411- ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/examples/412 413Paper from Linux-Kongress in Erlangen 2004.414- ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/pktgen_paper.pdf415 416Thanks to:417 418Grant Grundler for testing on IA-64 and parisc, Harald Welte, Lennert Buytenhek419Stephen Hemminger, Andi Kleen, Dave Miller and many others.420 421 422Good luck with the linux net-development.423