leandrof/linux
1
0
Fork 0
forked from Minki/linux
Code Issues Pull Requests Packages Projects Releases Wiki Activity
18a47e6d8a
linux/Documentation/networking/pktgen.txt
Jesper Dangaard Brouer 9ceb87fcea pktgen: document tuning for max NIC performance 
Using pktgen I'm seeing the ixgbe driver "push-back", due TX ring
running full.  Thus, the TX ring is artificially limiting pktgen.
(Diagnose via "ethtool -S", look for "tx_restart_queue" or "tx_busy"
counters.)

Using ixgbe, the real reason behind the TX ring running full, is due
to TX ring not being cleaned up fast enough. The ixgbe driver combines
TX+RX ring cleanups, and the cleanup interval is affected by the
ethtool --coalesce setting of parameter "rx-usecs".

Do not increase the default NIC TX ring buffer or default cleanup
interval.  Instead simply document that pktgen needs special NIC
tuning for maximum packet per sec performance.

Performance results with pktgen with clone_skb=100000.
TX ring size 512 (default), adjusting "rx-usecs":
 (Single CPU performance, E5-2630, ixgbe)
 - 3935002 pps - rx-usecs:  1 (irqs:  9346)
 - 5132350 pps - rx-usecs: 10 (irqs: 99157)
 - 5375111 pps - rx-usecs: 20 (irqs: 50154)
 - 5454050 pps - rx-usecs: 30 (irqs: 33872)
 - 5496320 pps - rx-usecs: 40 (irqs: 26197)
 - 5502510 pps - rx-usecs: 50 (irqs: 21527)

TX ring size adjusting (ethtool -G), "rx-usecs==1" (default):
 - 3935002 pps - tx-size:  512
 - 5354401 pps - tx-size:  768
 - 5356847 pps - tx-size: 1024
 - 5327595 pps - tx-size: 1536
 - 5356779 pps - tx-size: 2048
 - 5353438 pps - tx-size: 4096

Notice after commit 6f25cd47d (pktgen: fix xmit test for BQL enabled
devices) pktgen uses netif_xmit_frozen_or_drv_stopped() and ignores
the BQL "stack" pause (QUEUE_STATE_STACK_XOFF) flag.  This allow us to put
more pressure on the TX ring buffers.

It is the ixgbe_maybe_stop_tx() call that stops the transmits, and
pktgen respecting this in the call to netif_xmit_frozen_or_drv_stopped(txq).

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-07-01 15:50:23 -07:00

319 lines
9.8 KiB
Plaintext
Raw Blame History

HOWTO for the linux packet generator
------------------------------------
Date: 041221
Enable CONFIG_NET_PKTGEN to compile and build pktgen.o either in kernel
or as module. Module is preferred. insmod pktgen if needed. Once running
pktgen creates a thread on each CPU where each thread has affinity to its CPU.
Monitoring and controlling is done via /proc. Easiest to select a suitable
a sample script and configure.
On a dual CPU:
ps aux | grep pkt
root 129 0.3 0.0 0 0 ? SW 2003 523:20 [pktgen/0]
root 130 0.3 0.0 0 0 ? SW 2003 509:50 [pktgen/1]
For monitoring and control pktgen creates:
/proc/net/pktgen/pgctrl
/proc/net/pktgen/kpktgend_X
/proc/net/pktgen/ethX
Tuning NIC for max performance
==============================
The default NIC setting are (likely) not tuned for pktgen's artificial
overload type of benchmarking, as this could hurt the normal use-case.
Specifically increasing the TX ring buffer in the NIC:
# ethtool -G ethX tx 1024
A larger TX ring can improve pktgen's performance, while it can hurt
in the general case, 1) because the TX ring buffer might get larger
than the CPUs L1/L2 cache, 2) because it allow more queueing in the
NIC HW layer (which is bad for bufferbloat).
One should be careful to conclude, that packets/descriptors in the HW
TX ring cause delay. Drivers usually delay cleaning up the
ring-buffers (for various performance reasons), thus packets stalling
the TX ring, might just be waiting for cleanup.
This cleanup issues is specifically the case, for the driver ixgbe
(Intel 82599 chip). This driver (ixgbe) combine TX+RX ring cleanups,
and the cleanup interval is affected by the ethtool --coalesce setting
of parameter "rx-usecs".
For ixgbe use e.g "30" resulting in approx 33K interrupts/sec (1/30*10^6):
# ethtool -C ethX rx-usecs 30
Viewing threads
===============
/proc/net/pktgen/kpktgend_0
Name: kpktgend_0 max_before_softirq: 10000
Running:
Stopped: eth1
Result: OK: max_before_softirq=10000
Most important the devices assigned to thread. Note! A device can only belong
to one thread.
Viewing devices
===============
Parm section holds configured info. Current hold running stats.
Result is printed after run or after interruption. Example:
/proc/net/pktgen/eth1
Params: count 10000000 min_pkt_size: 60 max_pkt_size: 60
frags: 0 delay: 0 clone_skb: 1000000 ifname: eth1
flows: 0 flowlen: 0
dst_min: 10.10.11.2 dst_max:
src_min: src_max:
src_mac: 00:00:00:00:00:00 dst_mac: 00:04:23:AC:FD:82
udp_src_min: 9 udp_src_max: 9 udp_dst_min: 9 udp_dst_max: 9
src_mac_count: 0 dst_mac_count: 0
Flags:
Current:
pkts-sofar: 10000000 errors: 39664
started: 1103053986245187us stopped: 1103053999346329us idle: 880401us
seq_num: 10000011 cur_dst_mac_offset: 0 cur_src_mac_offset: 0
cur_saddr: 0x10a0a0a cur_daddr: 0x20b0a0a
cur_udp_dst: 9 cur_udp_src: 9
flows: 0
Result: OK: 13101142(c12220741+d880401) usec, 10000000 (60byte,0frags)
763292pps 390Mb/sec (390805504bps) errors: 39664
Configuring threads and devices
================================
This is done via the /proc interface easiest done via pgset in the scripts
Examples:
pgset "clone_skb 1" sets the number of copies of the same packet
pgset "clone_skb 0" use single SKB for all transmits
pgset "pkt_size 9014" sets packet size to 9014
pgset "frags 5" packet will consist of 5 fragments
pgset "count 200000" sets number of packets to send, set to zero
for continuous sends until explicitly stopped.
pgset "delay 5000" adds delay to hard_start_xmit(). nanoseconds
pgset "dst 10.0.0.1" sets IP destination address
(BEWARE! This generator is very aggressive!)
pgset "dst_min 10.0.0.1" Same as dst
pgset "dst_max 10.0.0.254" Set the maximum destination IP.
pgset "src_min 10.0.0.1" Set the minimum (or only) source IP.
pgset "src_max 10.0.0.254" Set the maximum source IP.
pgset "dst6 fec0::1" IPV6 destination address
pgset "src6 fec0::2" IPV6 source address
pgset "dstmac 00:00:00:00:00:00" sets MAC destination address
pgset "srcmac 00:00:00:00:00:00" sets MAC source address
pgset "queue_map_min 0" Sets the min value of tx queue interval
pgset "queue_map_max 7" Sets the max value of tx queue interval, for multiqueue devices
To select queue 1 of a given device,
use queue_map_min=1 and queue_map_max=1
pgset "src_mac_count 1" Sets the number of MACs we'll range through.
The 'minimum' MAC is what you set with srcmac.
pgset "dst_mac_count 1" Sets the number of MACs we'll range through.
The 'minimum' MAC is what you set with dstmac.
pgset "flag [name]" Set a flag to determine behaviour. Current flags
are: IPSRC_RND # IP source is random (between min/max)
IPDST_RND # IP destination is random
UDPSRC_RND, UDPDST_RND,
MACSRC_RND, MACDST_RND
TXSIZE_RND, IPV6,
MPLS_RND, VID_RND, SVID_RND
FLOW_SEQ,
QUEUE_MAP_RND # queue map random
QUEUE_MAP_CPU # queue map mirrors smp_processor_id()
UDPCSUM,
IPSEC # IPsec encapsulation (needs CONFIG_XFRM)
NODE_ALLOC # node specific memory allocation
pgset spi SPI_VALUE Set specific SA used to transform packet.
pgset "udp_src_min 9" set UDP source port min, If < udp_src_max, then
cycle through the port range.
pgset "udp_src_max 9" set UDP source port max.
pgset "udp_dst_min 9" set UDP destination port min, If < udp_dst_max, then
cycle through the port range.
pgset "udp_dst_max 9" set UDP destination port max.
pgset "mpls 0001000a,0002000a,0000000a" set MPLS labels (in this example
outer label=16,middle label=32,
inner label=0 (IPv4 NULL)) Note that
there must be no spaces between the
arguments. Leading zeros are required.
Do not set the bottom of stack bit,
that's done automatically. If you do
set the bottom of stack bit, that
indicates that you want to randomly
generate that address and the flag
MPLS_RND will be turned on. You
can have any mix of random and fixed
labels in the label stack.
pgset "mpls 0" turn off mpls (or any invalid argument works too!)
pgset "vlan_id 77" set VLAN ID 0-4095
pgset "vlan_p 3" set priority bit 0-7 (default 0)
pgset "vlan_cfi 0" set canonical format identifier 0-1 (default 0)
pgset "svlan_id 22" set SVLAN ID 0-4095
pgset "svlan_p 3" set priority bit 0-7 (default 0)
pgset "svlan_cfi 0" set canonical format identifier 0-1 (default 0)
pgset "vlan_id 9999" > 4095 remove vlan and svlan tags
pgset "svlan 9999" > 4095 remove svlan tag
pgset "tos XX" set former IPv4 TOS field (e.g. "tos 28" for AF11 no ECN, default 00)
pgset "traffic_class XX" set former IPv6 TRAFFIC CLASS (e.g. "traffic_class B8" for EF no ECN, default 00)
pgset stop aborts injection. Also, ^C aborts generator.
pgset "rate 300M" set rate to 300 Mb/s
pgset "ratep 1000000" set rate to 1Mpps
Example scripts
===============
A collection of small tutorial scripts for pktgen is in examples dir.
pktgen.conf-1-1 # 1 CPU 1 dev
pktgen.conf-1-2 # 1 CPU 2 dev
pktgen.conf-2-1 # 2 CPU's 1 dev
pktgen.conf-2-2 # 2 CPU's 2 dev
pktgen.conf-1-1-rdos # 1 CPU 1 dev w. route DoS
pktgen.conf-1-1-ip6 # 1 CPU 1 dev ipv6
pktgen.conf-1-1-ip6-rdos # 1 CPU 1 dev ipv6 w. route DoS
pktgen.conf-1-1-flows # 1 CPU 1 dev multiple flows.
Run in shell: ./pktgen.conf-X-Y It does all the setup including sending.
Interrupt affinity
===================
Note when adding devices to a specific CPU there good idea to also assign
/proc/irq/XX/smp_affinity so the TX-interrupts gets bound to the same CPU.
as this reduces cache bouncing when freeing skb's.
Enable IPsec
============
Default IPsec transformation with ESP encapsulation plus Transport mode
could be enabled by simply setting:
pgset "flag IPSEC"
pgset "flows 1"
To avoid breaking existing testbed scripts for using AH type and tunnel mode,
user could use "pgset spi SPI_VALUE" to specify which formal of transformation
to employ.
Current commands and configuration options
==========================================
** Pgcontrol commands:
start
stop
** Thread commands:
add_device
rem_device_all
max_before_softirq
** Device commands:
count
clone_skb
debug
frags
delay
src_mac_count
dst_mac_count
pkt_size
min_pkt_size
max_pkt_size
mpls
udp_src_min
udp_src_max
udp_dst_min
udp_dst_max
flag
IPSRC_RND
IPDST_RND
UDPSRC_RND
UDPDST_RND
MACSRC_RND
MACDST_RND
TXSIZE_RND
IPV6
MPLS_RND
VID_RND
SVID_RND
FLOW_SEQ
QUEUE_MAP_RND
QUEUE_MAP_CPU
UDPCSUM
IPSEC
NODE_ALLOC
dst_min
dst_max
src_min
src_max
dst_mac
src_mac
clear_counters
dst6
src6
flows
flowlen
rate
ratep
References:
ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/
ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/examples/
Paper from Linux-Kongress in Erlangen 2004.
ftp://robur.slu.se/pub/Linux/net-development/pktgen-testing/pktgen_paper.pdf
Thanks to:
Grant Grundler for testing on IA-64 and parisc, Harald Welte, Lennert Buytenhek
Stephen Hemminger, Andi Kleen, Dave Miller and many others.
Good luck with the linux net-development.
Reference in New Issue View Git Blame Copy Permalink