1 2 3 HOWTO for the linux packet generator 4 ------------------------------------ 5 6Enable CONFIG_NET_PKTGEN to compile and build pktgen either in-kernel 7or as a module. A module is preferred; modprobe pktgen if needed. Once 8running, pktgen creates a thread for each CPU with affinity to that CPU. 9Monitoring and controlling is done via /proc. It is easiest to select a 10suitable sample script and configure that. 11 12On a dual CPU: 13 14ps aux | grep pkt 15root 129 0.3 0.0 0 0 ? SW 2003 523:20 [pktgen/0] 16root 130 0.3 0.0 0 0 ? SW 2003 509:50 [pktgen/1] 17 18 19For monitoring and control pktgen creates: 20 /proc/net/pktgen/pgctrl 21 /proc/net/pktgen/kpktgend_X 22 /proc/net/pktgen/ethX 23 24 25Tuning NIC for max performance 26============================== 27 28The default NIC settings are (likely) not tuned for pktgen's artificial 29overload type of benchmarking, as this could hurt the normal use-case. 30 31Specifically increasing the TX ring buffer in the NIC: 32 # ethtool -G ethX tx 1024 33 34A larger TX ring can improve pktgen's performance, while it can hurt 35in the general case, 1) because the TX ring buffer might get larger 36than the CPU's L1/L2 cache, 2) because it allows more queueing in the 37NIC HW layer (which is bad for bufferbloat). 38 39One should hesitate to conclude that packets/descriptors in the HW 40TX ring cause delay. Drivers usually delay cleaning up the 41ring-buffers for various performance reasons, and packets stalling 42the TX ring might just be waiting for cleanup. 43 44This cleanup issue is specifically the case for the driver ixgbe 45(Intel 82599 chip). This driver (ixgbe) combines TX+RX ring cleanups, 46and the cleanup interval is affected by the ethtool --coalesce setting 47of parameter "rx-usecs". 48 49For ixgbe use e.g. "30" resulting in approx 33K interrupts/sec (1/30*10^6): 50 # ethtool -C ethX rx-usecs 30 51 52 53Viewing threads 54=============== 55/proc/net/pktgen/kpktgend_0 56Name: kpktgend_0 max_before_softirq: 10000 57Running: 58Stopped: eth1 59Result: OK: max_before_softirq=10000 60 61Most important are the devices assigned to the thread. Note that a 62device can only belong to one thread. 63 64 65Viewing devices 66=============== 67 68The Params section holds configured information. The Current section 69holds running statistics. The Result is printed after a run or after 70interruption. Example: 71 72/proc/net/pktgen/eth1 73 74Params: count 10000000 min_pkt_size: 60 max_pkt_size: 60 75 frags: 0 delay: 0 clone_skb: 1000000 ifname: eth1 76 flows: 0 flowlen: 0 77 dst_min: 10.10.11.2 dst_max: 78 src_min: src_max: 79 src_mac: 00:00:00:00:00:00 dst_mac: 00:04:23:AC:FD:82 80 udp_src_min: 9 udp_src_max: 9 udp_dst_min: 9 udp_dst_max: 9 81 src_mac_count: 0 dst_mac_count: 0 82 Flags: 83Current: 84 pkts-sofar: 10000000 errors: 39664 85 started: 1103053986245187us stopped: 1103053999346329us idle: 880401us 86 seq_num: 10000011 cur_dst_mac_offset: 0 cur_src_mac_offset: 0 87 cur_saddr: 0x10a0a0a cur_daddr: 0x20b0a0a 88 cur_udp_dst: 9 cur_udp_src: 9 89 flows: 0 90Result: OK: 13101142(c12220741+d880401) usec, 10000000 (60byte,0frags) 91 763292pps 390Mb/sec (390805504bps) errors: 39664 92 93Configuring threads and devices 94================================ 95This is done via the /proc interface, and most easily done via pgset 96as defined in the sample scripts. 97 98Examples: 99 100 pgset "clone_skb 1" sets the number of copies of the same packet 101 pgset "clone_skb 0" use single SKB for all transmits 102 pgset "burst 8" uses xmit_more API to queue 8 copies of the same 103 packet and update HW tx queue tail pointer once. 104 "burst 1" is the default 105 pgset "pkt_size 9014" sets packet size to 9014 106 pgset "frags 5" packet will consist of 5 fragments 107 pgset "count 200000" sets number of packets to send, set to zero 108 for continuous sends until explicitly stopped. 109 110 pgset "delay 5000" adds delay to hard_start_xmit(). nanoseconds 111 112 pgset "dst 10.0.0.1" sets IP destination address 113 (BEWARE! This generator is very aggressive!) 114 115 pgset "dst_min 10.0.0.1" Same as dst 116 pgset "dst_max 10.0.0.254" Set the maximum destination IP. 117 pgset "src_min 10.0.0.1" Set the minimum (or only) source IP. 118 pgset "src_max 10.0.0.254" Set the maximum source IP. 119 pgset "dst6 fec0::1" IPV6 destination address 120 pgset "src6 fec0::2" IPV6 source address 121 pgset "dstmac 00:00:00:00:00:00" sets MAC destination address 122 pgset "srcmac 00:00:00:00:00:00" sets MAC source address 123 124 pgset "queue_map_min 0" Sets the min value of tx queue interval 125 pgset "queue_map_max 7" Sets the max value of tx queue interval, for multiqueue devices 126 To select queue 1 of a given device, 127 use queue_map_min=1 and queue_map_max=1 128 129 pgset "src_mac_count 1" Sets the number of MACs we'll range through. 130 The 'minimum' MAC is what you set with srcmac. 131 132 pgset "dst_mac_count 1" Sets the number of MACs we'll range through. 133 The 'minimum' MAC is what you set with dstmac. 134 135 pgset "flag [name]" Set a flag to determine behaviour. Current flags 136 are: IPSRC_RND # IP source is random (between min/max) 137 IPDST_RND # IP destination is random 138 UDPSRC_RND, UDPDST_RND, 139 MACSRC_RND, MACDST_RND 140 TXSIZE_RND, IPV6, 141 MPLS_RND, VID_RND, SVID_RND 142 FLOW_SEQ, 143 QUEUE_MAP_RND # queue map random 144 QUEUE_MAP_CPU # queue map mirrors smp_processor_id() 145 UDPCSUM, 146 IPSEC # IPsec encapsulation (needs CONFIG_XFRM) 147 NODE_ALLOC # node specific memory allocation 148 149 pgset spi SPI_VALUE Set specific SA used to transform packet. 150 151 pgset "udp_src_min 9" set UDP source port min, If < udp_src_max, then 152 cycle through the port range. 153 154 pgset "udp_src_max 9" set UDP source port max. 155 pgset "udp_dst_min 9" set UDP destination port min, If < udp_dst_max, then 156 cycle through the port range. 157 pgset "udp_dst_max 9" set UDP destination port max. 158 159 pgset "mpls 0001000a,0002000a,0000000a" set MPLS labels (in this example 160 outer label=16,middle label=32, 161 inner label=0 (IPv4 NULL)) Note that 162 there must be no spaces between the 163 arguments. Leading zeros are required. 164 Do not set the bottom of stack bit, 165 that's done automatically. If you do 166 set the bottom of stack bit, that 167 indicates that you want to randomly 168 generate that address and the flag 169 MPLS_RND will be turned on. You 170 can have any mix of random and fixed 171 labels in the label stack. 172 173 pgset "mpls 0" turn off mpls (or any invalid argument works too!) 174 175 pgset "vlan_id 77" set VLAN ID 0-4095 176 pgset "vlan_p 3" set priority bit 0-7 (default 0) 177 pgset "vlan_cfi 0" set canonical format identifier 0-1 (default 0) 178 179 pgset "svlan_id 22" set SVLAN ID 0-4095 180 pgset "svlan_p 3" set priority bit 0-7 (default 0) 181 pgset "svlan_cfi 0" set canonical format identifier 0-1 (default 0) 182 183 pgset "vlan_id 9999" > 4095 remove vlan and svlan tags 184 pgset "svlan 9999" > 4095 remove svlan tag 185 186 187 pgset "tos XX" set former IPv4 TOS field (e.g. "tos 28" for AF11 no ECN, default 00) 188 pgset "traffic_class XX" set former IPv6 TRAFFIC CLASS (e.g. "traffic_class B8" for EF no ECN, default 00) 189 190 pgset stop aborts injection. Also, ^C aborts generator. 191 192 pgset "rate 300M" set rate to 300 Mb/s 193 pgset "ratep 1000000" set rate to 1Mpps 194 195Sample scripts 196============== 197 198A collection of small tutorial scripts for pktgen is in the 199samples/pktgen directory: 200 201pktgen.conf-1-1 # 1 CPU 1 dev 202pktgen.conf-1-2 # 1 CPU 2 dev 203pktgen.conf-2-1 # 2 CPU's 1 dev 204pktgen.conf-2-2 # 2 CPU's 2 dev 205pktgen.conf-1-1-rdos # 1 CPU 1 dev w. route DoS 206pktgen.conf-1-1-ip6 # 1 CPU 1 dev ipv6 207pktgen.conf-1-1-ip6-rdos # 1 CPU 1 dev ipv6 w. route DoS 208pktgen.conf-1-1-flows # 1 CPU 1 dev multiple flows. 209 210Run in shell: ./pktgen.conf-X-Y 211This does all the setup including sending. 212 213 214Interrupt affinity 215=================== 216Note that when adding devices to a specific CPU it is a good idea to 217also assign /proc/irq/XX/smp_affinity so that the TX interrupts are bound 218to the same CPU. This reduces cache bouncing when freeing skbs. 219 220Enable IPsec 221============ 222Default IPsec transformation with ESP encapsulation plus transport mode 223can be enabled by simply setting: 224 225pgset "flag IPSEC" 226pgset "flows 1" 227 228To avoid breaking existing testbed scripts for using AH type and tunnel mode, 229you can use "pgset spi SPI_VALUE" to specify which transformation mode 230to employ. 231 232 233Current commands and configuration options 234========================================== 235 236** Pgcontrol commands: 237 238start 239stop 240 241** Thread commands: 242 243add_device 244rem_device_all 245max_before_softirq 246 247 248** Device commands: 249 250count 251clone_skb 252debug 253 254frags 255delay 256 257src_mac_count 258dst_mac_count 259 260pkt_size 261min_pkt_size 262max_pkt_size 263 264mpls 265 266udp_src_min 267udp_src_max 268 269udp_dst_min 270udp_dst_max 271 272flag 273 IPSRC_RND 274 IPDST_RND 275 UDPSRC_RND 276 UDPDST_RND 277 MACSRC_RND 278 MACDST_RND 279 TXSIZE_RND 280 IPV6 281 MPLS_RND 282 VID_RND 283 SVID_RND 284 FLOW_SEQ 285 QUEUE_MAP_RND 286 QUEUE_MAP_CPU 287 UDPCSUM 288 IPSEC 289 NODE_ALLOC 290 291dst_min 292dst_max 293 294src_min 295src_max 296 297dst_mac 298src_mac 299 300clear_counters 301 302dst6 303src6 304 305flows 306flowlen 307 308rate 309ratep 310 311References: 312ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/ 313ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/examples/ 314 315Paper from Linux-Kongress in Erlangen 2004. 316ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/pktgen_paper.pdf 317 318Thanks to: 319Grant Grundler for testing on IA-64 and parisc, Harald Welte, Lennert Buytenhek 320Stephen Hemminger, Andi Kleen, Dave Miller and many others. 321 322 323Good luck with the linux net-development. 324