1 Linux* Base Driver for the Intel(R) PRO/10GbE Family of Adapters
2 ================================================================
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 is intended for 2.4.x kernels; it is
22 known to build properly on 2.4.x kernels through 2.4.18. Intel focused
23 testing on Intel architectures running kernels 2.4.18. This driver includes
24 support for Itanium(TM)-based systems.
26 For questions related to hardware requirements, refer to the documentation
27 supplied with your Intel PRO/10GbE adapter. All hardware requirements listed
28 apply to use with Linux.
34 The following Intel network adapters are compatible with the drivers in this
37 Controller Adapter Name Board IDs
38 ---------- ------------ ---------
40 82597EX Intel(R) PRO/10GbE LR Server Adapter A82505-xxx
43 To verify your Intel adapter is supported, find the board ID number on the
44 adapter. Look for a label that has a barcode and a number in the format
45 A12345-001. Match this to the list of numbers above.
47 For more information on how to identify your adapter, go to the Adapter &
50 http://support.intel.com/support/network/adapter/pro100/21397.htm
52 For the latest Intel network drivers for Linux, go to:
54 http://downloadfinder.intel.com/scripts-df/support_intel.asp
56 Command Line Parameters
57 =======================
59 If the driver is built as a module, the following optional parameters are
60 used by entering them on the command line with the modprobe or insmod command
63 modprobe ixgb [<option>=<VAL1>,<VAL2>,...]
65 insmod ixgb [<option>=<VAL1>,<VAL2>,...]
67 For example, with two PRO/10GbE PCI adapters, entering:
69 insmod ixgb TxDescriptors=80,128
71 loads the ixgb driver with 80 TX resources for the first adapter and 128 TX
72 resources for the second adapter.
74 The default value for each parameter is generally the recommended setting,
75 unless otherwise noted.
79 Valid Range: 0-3 (0=none, 1=Rx only, 2=Tx only, 3=Rx&Tx)
80 Default: Read from the EEPROM
81 If EEPROM is not detected, default is 3
82 This parameter controls the automatic generation(Tx) and response(Rx) to
83 Ethernet PAUSE frames.
88 This value is the number of receive descriptors allocated by the driver.
89 Increasing this value allows the driver to buffer more incoming packets.
90 Each descriptor is 16 bytes. A receive buffer is also allocated for
91 each descriptor and can be either 2048, 4056, 8192, or 16384 bytes,
92 depending on the MTU setting. When the MTU size is 1500 or less, the
93 receive buffer size is 2048 bytes. When the MTU is greater than 1500 the
94 receive buffer size will be either 4056, 8192, or 16384 bytes. The
95 maximum MTU size is 16114.
98 Valid Range: 0-65535 (0=off)
100 This value delays the generation of receive interrupts in units of
101 0.8192 microseconds. Receive interrupt reduction can improve CPU
102 efficiency if properly tuned for specific network traffic. Increasing
103 this value adds extra latency to frame reception and can end up
104 decreasing the throughput of TCP traffic. If the system is reporting
105 dropped receives, this value may be set too high, causing the driver to
106 run out of available receive descriptors.
111 This value is the number of transmit descriptors allocated by the driver.
112 Increasing this value allows the driver to queue more transmits. Each
113 descriptor is 16 bytes.
118 A value of '1' indicates that the driver should enable IP checksum
119 offload for received packets (both UDP and TCP) to the adapter hardware.
124 A value of '1' indicates that the driver should enable IP checksum
125 offload for transmitted packets (both UDP and TCP) to the adapter
129 Improving Performance
130 =====================
132 With the Intel PRO/10 GbE adapter, the default Linux configuration will very
133 likely limit the total available throughput artificially. There is a set of
134 things that when applied together increase the ability of Linux to transmit
135 and receive data. The following enhancements were originally acquired from
136 settings published at http://www.spec.org/web99 for various submitted results
139 NOTE: These changes are only suggestions, and serve as a starting point for
140 tuning your network performance.
142 The changes are made in three major ways, listed in order of greatest effect:
143 - Use ifconfig to modify the mtu (maximum transmission unit) and the txqueuelen
145 - Use sysctl to modify /proc parameters (essentially kernel tuning)
146 - Use setpci to modify the MMRBC field in PCI-X configuration space to increase
147 transmit burst lengths on the bus.
149 NOTE: setpci modifies the adapter's configuration registers to allow it to read
150 up to 4k bytes at a time (for transmits). However, for some systems the
151 behavior after modifying this register may be undefined (possibly errors of some
152 kind). A power-cycle, hard reset or explicitly setting the e6 register back to
153 22 (setpci -d 8086:1048 e6.b=22) may be required to get back to a stable
156 - COPY these lines and paste them into ixgb_perf.sh:
158 echo "configuring network performance , edit this file to change the interface"
159 # set mmrbc to 4k reads, modify only Intel 10GbE device IDs
160 setpci -d 8086:1048 e6.b=2e
161 # set the MTU (max transmission unit) - it requires your switch and clients to change too!
163 # your ixgb adapter should be loaded as eth1 for this to work, change if needed
164 ifconfig eth1 mtu 9000 txqueuelen 1000 up
165 # call the sysctl utility to modify /proc/sys entries
166 sysctl -p ./sysctl_ixgb.conf
169 - COPY these lines and paste them into sysctl_ixgb.conf:
170 # some of the defaults may be different for your kernel
171 # call this file with sysctl -p <this file>
172 # these are just suggested values that worked well to increase throughput in
173 # several network benchmark tests, your mileage may vary
175 ### IPV4 specific settings
176 net.ipv4.tcp_timestamps = 0 # turns TCP timestamp support off, default 1, reduces CPU use
177 net.ipv4.tcp_sack = 0 # turn SACK support off, default on
178 # on systems with a VERY fast bus -> memory interface this is the big gainer
179 net.ipv4.tcp_rmem = 10000000 10000000 10000000 # sets min/default/max TCP read buffer, default 4096 87380 174760
180 net.ipv4.tcp_wmem = 10000000 10000000 10000000 # sets min/pressure/max TCP write buffer, default 4096 16384 131072
181 net.ipv4.tcp_mem = 10000000 10000000 10000000 # sets min/pressure/max TCP buffer space, default 31744 32256 32768
183 ### CORE settings (mostly for socket and UDP effect)
184 net.core.rmem_max = 524287 # maximum receive socket buffer size, default 131071
185 net.core.wmem_max = 524287 # maximum send socket buffer size, default 131071
186 net.core.rmem_default = 524287 # default receive socket buffer size, default 65535
187 net.core.wmem_default = 524287 # default send socket buffer size, default 65535
188 net.core.optmem_max = 524287 # maximum amount of option memory buffers, default 10240
189 net.core.netdev_max_backlog = 300000 # number of unprocessed input packets before kernel starts dropping them, default 300
190 - END sysctl_ixgb.conf
192 Edit the ixgb_perf.sh script if necessary to change eth1 to whatever interface
193 your ixgb driver is using.
195 NOTE: Unless these scripts are added to the boot process, these changes will
196 only last only until the next system reboot.
199 Resolving Slow UDP Traffic
200 --------------------------
202 If your server does not seem to be able to receive UDP traffic as fast as it
203 can receive TCP traffic, it could be because Linux, by default, does not set
204 the network stack buffers as large as they need to be to support high UDP
205 transfer rates. One way to alleviate this problem is to allow more memory to
206 be used by the IP stack to store incoming data.
208 For instance, use the commands:
209 sysctl -w net.core.rmem_max=262143
211 sysctl -w net.core.rmem_default=262143
212 to increase the read buffer memory max and default to 262143 (256k - 1) from
213 defaults of max=131071 (128k - 1) and default=65535 (64k - 1). These variables
214 will increase the amount of memory used by the network stack for receives, and
215 can be increased significantly more if necessary for your application.
220 For general information and support, go to the Intel support website at:
222 http://support.intel.com
224 If an issue is identified with the released source code on the supported
225 kernel with a supported adapter, email the specific information related to
226 the issue to linux.nics@intel.com.