1 Linux Base Driver for 10 Gigabit Intel(R) Network Connection
2 =============================================================
11 - Identifying Your Adapter
12 - Building and Installation
13 - Command Line Parameters
14 - Improving Performance
15 - Additional Configurations
16 - Known Issues/Troubleshooting
24 This file describes the ixgb Linux Base Driver for the 10 Gigabit Intel(R)
25 Network Connection. This driver includes support for Itanium(R)2-based
28 For questions related to hardware requirements, refer to the documentation
29 supplied with your 10 Gigabit adapter. All hardware requirements listed apply
32 The following features are available in this kernel:
34 - Channel Bonding (teaming)
37 Channel Bonding documentation can be found in the Linux kernel source:
38 /Documentation/networking/bonding.txt
40 The driver information previously displayed in the /proc filesystem is not
41 supported in this release. Alternatively, you can use ethtool (version 1.6
42 or later), lspci, and ifconfig to obtain the same information.
44 Instructions on updating ethtool can be found in the section "Additional
45 Configurations" later in this document.
48 Identifying Your Adapter
49 ========================
51 The following Intel network adapters are compatible with the drivers in this
54 Controller Adapter Name Physical Layer
55 ---------- ------------ --------------
56 82597EX Intel(R) PRO/10GbE LR/SR/CX4 10G Base-LR (1310 nm optical fiber)
57 Server Adapters 10G Base-SR (850 nm optical fiber)
58 10G Base-CX4(twin-axial copper cabling)
60 For more information on how to identify your adapter, go to the Adapter &
63 http://support.intel.com/support/network/sb/CS-012904.htm
66 Building and Installation
67 =========================
69 select m for "Intel(R) PRO/10GbE support" located at:
72 -> Network device support (NETDEVICES [=y])
73 -> Ethernet (10000 Mbit) (NETDEV_10000 [=y])
74 1. make modules && make modules_install
78 modprobe ixgb <parameter>=<value>
80 The insmod command can be used if the full
81 path to the driver module is specified. For example:
83 insmod /lib/modules/<KERNEL VERSION>/kernel/drivers/net/ixgb/ixgb.ko
85 With 2.6 based kernels also make sure that older ixgb drivers are
86 removed from the kernel, before loading the new module:
88 rmmod ixgb; modprobe ixgb
90 3. Assign an IP address to the interface by entering the following, where
91 x is the interface number:
93 ifconfig ethx <IP_address>
95 4. Verify that the interface works. Enter the following, where <IP_address>
96 is the IP address for another machine on the same subnet as the interface
102 Command Line Parameters
103 =======================
105 If the driver is built as a module, the following optional parameters are
106 used by entering them on the command line with the modprobe command using
109 modprobe ixgb [<option>=<VAL1>,<VAL2>,...]
111 For example, with two 10GbE PCI adapters, entering:
113 modprobe ixgb TxDescriptors=80,128
115 loads the ixgb driver with 80 TX resources for the first adapter and 128 TX
116 resources for the second adapter.
118 The default value for each parameter is generally the recommended setting,
119 unless otherwise noted.
122 Valid Range: 0-3 (0=none, 1=Rx only, 2=Tx only, 3=Rx&Tx)
123 Default: Read from the EEPROM
124 If EEPROM is not detected, default is 1
125 This parameter controls the automatic generation(Tx) and response(Rx) to
126 Ethernet PAUSE frames. There are hardware bugs associated with enabling
127 Tx flow control so beware.
132 This value is the number of receive descriptors allocated by the driver.
133 Increasing this value allows the driver to buffer more incoming packets.
134 Each descriptor is 16 bytes. A receive buffer is also allocated for
135 each descriptor and can be either 2048, 4056, 8192, or 16384 bytes,
136 depending on the MTU setting. When the MTU size is 1500 or less, the
137 receive buffer size is 2048 bytes. When the MTU is greater than 1500 the
138 receive buffer size will be either 4056, 8192, or 16384 bytes. The
139 maximum MTU size is 16114.
142 Valid Range: 0-65535 (0=off)
144 This value delays the generation of receive interrupts in units of
145 0.8192 microseconds. Receive interrupt reduction can improve CPU
146 efficiency if properly tuned for specific network traffic. Increasing
147 this value adds extra latency to frame reception and can end up
148 decreasing the throughput of TCP traffic. If the system is reporting
149 dropped receives, this value may be set too high, causing the driver to
150 run out of available receive descriptors.
155 This value is the number of transmit descriptors allocated by the driver.
156 Increasing this value allows the driver to queue more transmits. Each
157 descriptor is 16 bytes.
162 A value of '1' indicates that the driver should enable IP checksum
163 offload for received packets (both UDP and TCP) to the adapter hardware.
166 Improving Performance
167 =====================
169 With the 10 Gigabit server adapters, the default Linux configuration will
170 very likely limit the total available throughput artificially. There is a set
171 of configuration changes that, when applied together, will increase the ability
172 of Linux to transmit and receive data. The following enhancements were
173 originally acquired from settings published at http://www.spec.org/web99/ for
174 various submitted results using Linux.
176 NOTE: These changes are only suggestions, and serve as a starting point for
177 tuning your network performance.
179 The changes are made in three major ways, listed in order of greatest effect:
180 - Use ifconfig to modify the mtu (maximum transmission unit) and the txqueuelen
182 - Use sysctl to modify /proc parameters (essentially kernel tuning)
183 - Use setpci to modify the MMRBC field in PCI-X configuration space to increase
184 transmit burst lengths on the bus.
186 NOTE: setpci modifies the adapter's configuration registers to allow it to read
187 up to 4k bytes at a time (for transmits). However, for some systems the
188 behavior after modifying this register may be undefined (possibly errors of
189 some kind). A power-cycle, hard reset or explicitly setting the e6 register
190 back to 22 (setpci -d 8086:1a48 e6.b=22) may be required to get back to a
191 stable configuration.
193 - COPY these lines and paste them into ixgb_perf.sh:
195 echo "configuring network performance , edit this file to change the interface
196 or device ID of 10GbE card"
197 # set mmrbc to 4k reads, modify only Intel 10GbE device IDs
198 # replace 1a48 with appropriate 10GbE device's ID installed on the system,
200 setpci -d 8086:1a48 e6.b=2e
201 # set the MTU (max transmission unit) - it requires your switch and clients
204 # your ixgb adapter should be loaded as eth1 for this to work, change if needed
205 ifconfig eth1 mtu 9000 txqueuelen 1000 up
206 # call the sysctl utility to modify /proc/sys entries
207 sysctl -p ./sysctl_ixgb.conf
210 - COPY these lines and paste them into sysctl_ixgb.conf:
211 # some of the defaults may be different for your kernel
212 # call this file with sysctl -p <this file>
213 # these are just suggested values that worked well to increase throughput in
214 # several network benchmark tests, your mileage may vary
216 ### IPV4 specific settings
217 # turn TCP timestamp support off, default 1, reduces CPU use
218 net.ipv4.tcp_timestamps = 0
219 # turn SACK support off, default on
220 # on systems with a VERY fast bus -> memory interface this is the big gainer
221 net.ipv4.tcp_sack = 0
222 # set min/default/max TCP read buffer, default 4096 87380 174760
223 net.ipv4.tcp_rmem = 10000000 10000000 10000000
224 # set min/pressure/max TCP write buffer, default 4096 16384 131072
225 net.ipv4.tcp_wmem = 10000000 10000000 10000000
226 # set min/pressure/max TCP buffer space, default 31744 32256 32768
227 net.ipv4.tcp_mem = 10000000 10000000 10000000
229 ### CORE settings (mostly for socket and UDP effect)
230 # set maximum receive socket buffer size, default 131071
231 net.core.rmem_max = 524287
232 # set maximum send socket buffer size, default 131071
233 net.core.wmem_max = 524287
234 # set default receive socket buffer size, default 65535
235 net.core.rmem_default = 524287
236 # set default send socket buffer size, default 65535
237 net.core.wmem_default = 524287
238 # set maximum amount of option memory buffers, default 10240
239 net.core.optmem_max = 524287
240 # set number of unprocessed input packets before kernel starts dropping them; default 300
241 net.core.netdev_max_backlog = 300000
242 - END sysctl_ixgb.conf
244 Edit the ixgb_perf.sh script if necessary to change eth1 to whatever interface
245 your ixgb driver is using and/or replace '1a48' with appropriate 10GbE device's
246 ID installed on the system.
248 NOTE: Unless these scripts are added to the boot process, these changes will
249 only last only until the next system reboot.
252 Resolving Slow UDP Traffic
253 --------------------------
254 If your server does not seem to be able to receive UDP traffic as fast as it
255 can receive TCP traffic, it could be because Linux, by default, does not set
256 the network stack buffers as large as they need to be to support high UDP
257 transfer rates. One way to alleviate this problem is to allow more memory to
258 be used by the IP stack to store incoming data.
260 For instance, use the commands:
261 sysctl -w net.core.rmem_max=262143
263 sysctl -w net.core.rmem_default=262143
264 to increase the read buffer memory max and default to 262143 (256k - 1) from
265 defaults of max=131071 (128k - 1) and default=65535 (64k - 1). These variables
266 will increase the amount of memory used by the network stack for receives, and
267 can be increased significantly more if necessary for your application.
270 Additional Configurations
271 =========================
273 Configuring the Driver on Different Distributions
274 -------------------------------------------------
275 Configuring a network driver to load properly when the system is started is
276 distribution dependent. Typically, the configuration process involves adding
277 an alias line to /etc/modprobe.conf as well as editing other system startup
278 scripts and/or configuration files. Many popular Linux distributions ship
279 with tools to make these changes for you. To learn the proper way to
280 configure a network device for your system, refer to your distribution
281 documentation. If during this process you are asked for the driver or module
282 name, the name for the Linux Base Driver for the Intel 10GbE Family of
285 Viewing Link Messages
286 ---------------------
287 Link messages will not be displayed to the console if the distribution is
288 restricting system messages. In order to see network driver link messages on
289 your console, set dmesg to eight by entering the following:
293 NOTE: This setting is not saved across reboots.
298 The driver supports Jumbo Frames for all adapters. Jumbo Frames support is
299 enabled by changing the MTU to a value larger than the default of 1500.
300 The maximum value for the MTU is 16114. Use the ifconfig command to
301 increase the MTU size. For example:
303 ifconfig ethx mtu 9000 up
305 The maximum MTU setting for Jumbo Frames is 16114. This value coincides
306 with the maximum Jumbo Frames size of 16128.
311 The driver utilizes the ethtool interface for driver configuration and
312 diagnostics, as well as displaying statistical information. Ethtool
313 version 1.6 or later is required for this functionality.
315 The latest release of ethtool can be found from
316 http://sourceforge.net/projects/gkernel
318 NOTE: Ethtool 1.6 only supports a limited set of ethtool options. Support
319 for a more complete ethtool feature set can be enabled by upgrading
320 to the latest version.
326 NAPI (Rx polling mode) is supported in the ixgb driver. NAPI is enabled
327 or disabled based on the configuration of the kernel. see CONFIG_IXGB_NAPI
329 See www.cyberus.ca/~hadi/usenix-paper.tgz for more information on NAPI.
332 Known Issues/Troubleshooting
333 ============================
335 NOTE: After installing the driver, if your Intel Network Connection is not
336 working, verify in the "In This Release" section of the readme that you have
337 installed the correct driver.
339 Intel(R) PRO/10GbE CX4 Server Adapter Cable Interoperability Issue with
340 Fujitsu XENPAK Module in SmartBits Chassis
341 ---------------------------------------------------------------------
342 Excessive CRC errors may be observed if the Intel(R) PRO/10GbE CX4
343 Server adapter is connected to a Fujitsu XENPAK CX4 module in a SmartBits
344 chassis using 15 m/24AWG cable assemblies manufactured by Fujitsu or Leoni.
345 The CRC errors may be received either by the Intel(R) PRO/10GbE CX4
346 Server adapter or the SmartBits. If this situation occurs using a different
347 cable assembly may resolve the issue.
349 CX4 Server Adapter Cable Interoperability Issues with HP Procurve 3400cl
351 ------------------------------------------------------------------------
352 Excessive CRC errors may be observed if the Intel(R) PRO/10GbE CX4 Server
353 adapter is connected to an HP Procurve 3400cl switch port using short cables
354 (1 m or shorter). If this situation occurs, using a longer cable may resolve
357 Excessive CRC errors may be observed using Fujitsu 24AWG cable assemblies that
358 Are 10 m or longer or where using a Leoni 15 m/24AWG cable assembly. The CRC
359 errors may be received either by the CX4 Server adapter or at the switch. If
360 this situation occurs, using a different cable assembly may resolve the issue.
363 Jumbo Frames System Requirement
364 -------------------------------
365 Memory allocation failures have been observed on Linux systems with 64 MB
366 of RAM or less that are running Jumbo Frames. If you are using Jumbo
367 Frames, your system may require more than the advertised minimum
368 requirement of 64 MB of system memory.
371 Performance Degradation with Jumbo Frames
372 -----------------------------------------
373 Degradation in throughput performance may be observed in some Jumbo frames
374 environments. If this is observed, increasing the application's socket buffer
375 size and/or increasing the /proc/sys/net/ipv4/tcp_*mem entry values may help.
376 See the specific application manual and /usr/src/linux*/Documentation/
377 networking/ip-sysctl.txt for more details.
380 Allocating Rx Buffers when Using Jumbo Frames
381 ---------------------------------------------
382 Allocating Rx buffers when using Jumbo Frames on 2.6.x kernels may fail if
383 the available memory is heavily fragmented. This issue may be seen with PCI-X
384 adapters or with packet split disabled. This can be reduced or eliminated
385 by changing the amount of available memory for receive buffer allocation, by
386 increasing /proc/sys/vm/min_free_kbytes.
389 Multiple Interfaces on Same Ethernet Broadcast Network
390 ------------------------------------------------------
391 Due to the default ARP behavior on Linux, it is not possible to have
392 one system on two IP networks in the same Ethernet broadcast domain
393 (non-partitioned switch) behave as expected. All Ethernet interfaces
394 will respond to IP traffic for any IP address assigned to the system.
395 This results in unbalanced receive traffic.
397 If you have multiple interfaces in a server, do either of the following:
399 - Turn on ARP filtering by entering:
400 echo 1 > /proc/sys/net/ipv4/conf/all/arp_filter
402 - Install the interfaces in separate broadcast domains - either in
403 different switches or in a switch partitioned to VLANs.
406 UDP Stress Test Dropped Packet Issue
407 --------------------------------------
408 Under small packets UDP stress test with 10GbE driver, the Linux system
409 may drop UDP packets due to the fullness of socket buffers. You may want
410 to change the driver's Flow Control variables to the minimum value for
411 controlling packet reception.
414 Tx Hangs Possible Under Stress
415 ------------------------------
416 Under stress conditions, if TX hangs occur, turning off TSO
417 "ethtool -K eth0 tso off" may resolve the problem.
423 For general information, go to the Intel support website at:
425 http://support.intel.com
427 or the Intel Wired Networking project hosted by Sourceforge at:
429 http://sourceforge.net/projects/e1000
431 If an issue is identified with the released source code on the supported
432 kernel with a supported adapter, email the specific information related
433 to the issue to e1000-devel@lists.sf.net