1 Linux* Base Driver for the Intel(R) PRO/10GbE Family of Adapters
2 ================================================================
11 - Identifying Your Adapter
12 - Command Line Parameters
13 - Improving Performance
20 This file describes the Linux* Base Driver for the Intel(R) PRO/10GbE Family
21 of Adapters, version 1.0.x. This driver includes support for Itanium(TM)2 and
24 For questions related to hardware requirements, refer to the documentation
25 supplied with your Intel PRO/10GbE adapter. All hardware requirements listed
26 apply to use with Linux.
28 Identifying Your Adapter
29 ========================
31 To verify your Intel adapter is supported, find the board ID number on the
32 adapter. Look for a label that has a barcode and a number in the format
35 Use the above information and the Adapter & Driver ID Guide at:
37 http://support.intel.com/support/network/adapter/pro100/21397.htm
39 For the latest Intel network drivers for Linux, go to:
41 http://downloadfinder.intel.com/scripts-df/support_intel.asp
43 Command Line Parameters
44 =======================
46 If the driver is built as a module, the following optional parameters are
47 used by entering them on the command line with the modprobe or insmod command
50 modprobe ixgb [<option>=<VAL1>,<VAL2>,...]
52 insmod ixgb [<option>=<VAL1>,<VAL2>,...]
54 For example, with two PRO/10GbE PCI adapters, entering:
56 insmod ixgb TxDescriptors=80,128
58 loads the ixgb driver with 80 TX resources for the first adapter and 128 TX
59 resources for the second adapter.
61 The default value for each parameter is generally the recommended setting,
62 unless otherwise noted. Also, if the driver is statically built into the
63 kernel, the driver is loaded with the default values for all the parameters.
64 Ethtool can be used to change some of the parameters at runtime.
67 Valid Range: 0-3 (0=none, 1=Rx only, 2=Tx only, 3=Rx&Tx)
68 Default: Read from the EEPROM
69 If EEPROM is not detected, default is 3
70 This parameter controls the automatic generation(Tx) and response(Rx) to
71 Ethernet PAUSE frames.
76 This value is the number of receive descriptors allocated by the driver.
77 Increasing this value allows the driver to buffer more incoming packets.
78 Each descriptor is 16 bytes. A receive buffer is also allocated for
79 each descriptor and can be either 2048, 4056, 8192, or 16384 bytes,
80 depending on the MTU setting. When the MTU size is 1500 or less, the
81 receive buffer size is 2048 bytes. When the MTU is greater than 1500 the
82 receive buffer size will be either 4056, 8192, or 16384 bytes. The
83 maximum MTU size is 16114.
86 Valid Range: 0-65535 (0=off)
88 This value delays the generation of receive interrupts in units of
89 0.8192 microseconds. Receive interrupt reduction can improve CPU
90 efficiency if properly tuned for specific network traffic. Increasing
91 this value adds extra latency to frame reception and can end up
92 decreasing the throughput of TCP traffic. If the system is reporting
93 dropped receives, this value may be set too high, causing the driver to
94 run out of available receive descriptors.
99 This value is the number of transmit descriptors allocated by the driver.
100 Increasing this value allows the driver to queue more transmits. Each
101 descriptor is 16 bytes.
106 A value of '1' indicates that the driver should enable IP checksum
107 offload for received packets (both UDP and TCP) to the adapter hardware.
112 A value of '1' indicates that the driver should enable IP checksum
113 offload for transmitted packets (both UDP and TCP) to the adapter
116 Improving Performance
117 =====================
119 With the Intel PRO/10 GbE adapter, the default Linux configuration will very
120 likely limit the total available throughput artificially. There is a set of
121 things that when applied together increase the ability of Linux to transmit
122 and receive data. The following enhancements were originally acquired from
123 settings published at http://www.spec.org/web99 for various submitted results
126 NOTE: These changes are only suggestions, and serve as a starting point for
127 tuning your network performance.
129 The changes are made in three major ways, listed in order of greatest effect:
130 - Use ifconfig to modify the mtu (maximum transmission unit) and the txqueuelen
132 - Use sysctl to modify /proc parameters (essentially kernel tuning)
133 - Use setpci to modify the MMRBC field in PCI-X configuration space to increase
134 transmit burst lengths on the bus.
136 NOTE: setpci modifies the adapter's configuration registers to allow it to read
137 up to 4k bytes at a time (for transmits). However, for some systems the
138 behavior after modifying this register may be undefined (possibly errors of some
139 kind). A power-cycle, hard reset or explicitly setting the e6 register back to
140 22 (setpci -d 8086:1048 e6.b=22) may be required to get back to a stable
143 - COPY these lines and paste them into ixgb_perf.sh:
145 echo "configuring network performance , edit this file to change the interface"
146 # set mmrbc to 4k reads, modify only Intel 10GbE device IDs
147 setpci -d 8086:1048 e6.b=2e
148 # set the MTU (max transmission unit) - it requires your switch and clients to change too!
150 # your ixgb adapter should be loaded as eth1 for this to work, change if needed
151 ifconfig eth1 mtu 9000 txqueuelen 1000 up
152 # call the sysctl utility to modify /proc/sys entries
153 sysctl -p ./sysctl_ixgb.conf
156 - COPY these lines and paste them into sysctl_ixgb.conf:
157 # some of the defaults may be different for your kernel
158 # call this file with sysctl -p <this file>
159 # these are just suggested values that worked well to increase throughput in
160 # several network benchmark tests, your mileage may vary
162 ### IPV4 specific settings
163 net.ipv4.tcp_timestamps = 0 # turns TCP timestamp support off, default 1, reduces CPU use
164 net.ipv4.tcp_sack = 0 # turn SACK support off, default on
165 # on systems with a VERY fast bus -> memory interface this is the big gainer
166 net.ipv4.tcp_rmem = 10000000 10000000 10000000 # sets min/default/max TCP read buffer, default 4096 87380 174760
167 net.ipv4.tcp_wmem = 10000000 10000000 10000000 # sets min/pressure/max TCP write buffer, default 4096 16384 131072
168 net.ipv4.tcp_mem = 10000000 10000000 10000000 # sets min/pressure/max TCP buffer space, default 31744 32256 32768
170 ### CORE settings (mostly for socket and UDP effect)
171 net.core.rmem_max = 524287 # maximum receive socket buffer size, default 131071
172 net.core.wmem_max = 524287 # maximum send socket buffer size, default 131071
173 net.core.rmem_default = 524287 # default receive socket buffer size, default 65535
174 net.core.wmem_default = 524287 # default send socket buffer size, default 65535
175 net.core.optmem_max = 524287 # maximum amount of option memory buffers, default 10240
176 net.core.netdev_max_backlog = 300000 # number of unprocessed input packets before kernel starts dropping them, default 300
177 - END sysctl_ixgb.conf
179 Edit the ixgb_perf.sh script if necessary to change eth1 to whatever interface
180 your ixgb driver is using.
182 NOTE: Unless these scripts are added to the boot process, these changes will
183 only last only until the next system reboot.
186 Resolving Slow UDP Traffic
187 --------------------------
189 If your server does not seem to be able to receive UDP traffic as fast as it
190 can receive TCP traffic, it could be because Linux, by default, does not set
191 the network stack buffers as large as they need to be to support high UDP
192 transfer rates. One way to alleviate this problem is to allow more memory to
193 be used by the IP stack to store incoming data.
195 For instance, use the commands:
196 sysctl -w net.core.rmem_max=262143
198 sysctl -w net.core.rmem_default=262143
199 to increase the read buffer memory max and default to 262143 (256k - 1) from
200 defaults of max=131071 (128k - 1) and default=65535 (64k - 1). These variables
201 will increase the amount of memory used by the network stack for receives, and
202 can be increased significantly more if necessary for your application.
207 For general information and support, go to the Intel support website at:
209 http://support.intel.com
211 If an issue is identified with the released source code on the supported
212 kernel with a supported adapter, email the specific information related to
213 the issue to linux.nics@intel.com.