April 7, 2010
In the 2.4.x kernel, qdiscs were introduced. The rationale behind this effort was to provide QoS in software, as hardware did not provide the necessary interfaces to support it. In 2.6.23, Intel introduced the idea of multiqueue support into the qdisc layer. This provides a mechanism to map the software queues in the qdisc structure into multiple hardware queues in underlying devices. In the case of Intel adapters, this mechanism is used to map qdisc queues onto the queues within the hardware controllers.
Within the Data Center, the perception is that traditional Ethernet is subject to high latency and is prone to losing frames, rendering it unacceptable for storage applications
In an effort to address these issues, Intel and a host of industry leaders have been working on addressing these problems. Specifically, within the IEEE 802.1 standards body there are a number of task forces working on enhancements to address these concerns. Listed below are the applicable standards:
Enhanced Transmission Selection: IEEE 802.1Qaz Lossless Traffic Class Priority Flow Control: IEEE 802.1Qbb Congestion Notification: IEEE 802.1Qau DCB Capability exchange protocol (DCBX): IEEE 802.1Qaz
The software solution that is being released represents Intel's implementation of these efforts. It is worth noting that many of these standards have not been ratified - this is a pre-standards release, so users are advised to check Sourceforge often. While we have worked with some of the major ecosystem vendors in validating this release, there are many vendors which still have solutions in development. As these solutions become available and standards get ratified, we will work with ecosystem partners and the standards body to ensure that the Intel solution works as expected.
lldpad
lldptool
dcbtool
Required configuration options:
From make menuconfig a. In Networking support -> Networking options, enable Data Center Bridging b. In Networking support -> Networking options -> QoS and/or fair queuing, enable: Hardware Multiqueue-aware Multi Band Queuing (MULTIQ) Multi Band Priority Queueing (PRIO) Elementary classification ( BASIC ) Universal 32bit comparisons w/ hashing (U32) Extended Matches and make sure U32 key is selected Actions -> SKB Editing c. To enable ixgbe driver support for DCB, In Device Drivers -> Network device support -> Ethernet (10000 Mbit) enable Intel(R) 10GbE PCI Express adapters support and the Data Center Bridging (DCB) Support option
Build the kernel.
lldpad requires updated kernel header files to compile. Create a link from /usr/include/linux to the location of the kernel source include files. For example: ln -s /usr/src/kernels/linux-2.x.xx.x/include/linux /usr/include/linux
Download iproute2 from the web. See: http://www.linuxfoundation.org/en/Net:Iproute2 Please ensure that you use the version that corresponds to the kernel version that you are using. Follow the build/installation instructions in the README. Typically, the commands: ./configure; make; make install will work.
Download the latest version of the lldpad-x.y.z tarball from the e1000 project in Sourceforge and untar it.
Check to see that the following libraries are installed; install them if they are not present.
libtool; libconfig-devel; libnl-devel; readline-devel
Go into the lldpad-x.y.z directory and run the following commands
./bootstrap.sh; ./configure --prefix=/usr; make; make install
This will build and install 'lldpad', 'lldptool' and 'dcbtool' to /usr/sbin, make the '/var/lib/lldpad' directory (default location of the lldpad.conf file) and setup lldpad to run as a system service using the chkconfig program. Verify that the lldpad service is working as expected with the 'service lldpad status' command. If the service is not on, issue the command 'service lldpad start'
Note (known issue): if the --prefix option is not supplied to the configure script, then the default install location of 'lldpad', 'lldptool' and 'dcbtool' is /usr/local/sbin. This does not match the 'lldpad' script installed in /etc/init.d, which is coded to start lldpad from /usr/sbin/lldpad.
lldpad will create the lldpad.conf file if it does not exist.
For development purposes, 'lldpad' can be run directly from the build directory. To run the unit tests associated with openlldpd, execute:
./bootstrap.sh; ./configure; make check
lldpad has the following command line options: -h show usage information -f configfile use the specified file as the config file instead of the default file - /var/lib/lldpad/lldpad.conf -d run lldpad as a daemon -v show lldpad version -k terminate current running lldpad -s remove lldpad state records
Load the ixgbe module.
Verify lldpad service is functional. If lldpad was installed, do "service lldpad status" to check, "service lldpad start" to start. Or, run "lldpad -d" from the command line to start.
Enable DCB on the selected ixgbe port: dcbtool sc ethX dcb on
The dcbtool command can be used to query and change the DCB configuration (ie., various percentages to different queues). Use dcbtool -h to see a list of options.
lldpad and dcbtool can be used to configure a DCB capable driver, such as the ixgbe driver, which supports the rtnetlink DCB interface. Once the DCB features are configured, the next step is to classify traffic to be identified with an 802.1p priority and the associated DCB features. This can be done using a couple different methods.
Set up the multi-queue qdisc for the selected interface:
Setting the queue_mapping in a TC filter allows the ixgbe driver to classify a packet into a queue. Here is an example of a filter to cause iSCSI traffic (running on the well known port number 3260) to be tagged with priority 4.
0xffff action skbedit queue_mapping 4
0xffff action skbedit queue_mapping 4
Here is an example that sets up a filter based on EtherType. In this example, the EtherType is 0x8906.
'cmp(u16 at 12 layer 1 mask 0xffff eq 35078)' action skbedit queue_mapping 3
...
Note: since DCB features use the 802.1p priority in the VLAN tag to distinguish traffic, it is expected that traffic in a DCB network will need to be on a tagged VLAN.
To test in a back-to-back setup, use the following tc commands to setup the qdisc and filters for TCP ports 5000 through 5007. Then use a tool, such as iperf, to generate UDP or TCP traffic on ports 5000-5007.
Statistics for each queue of the ixgbe driver can be checked using the ethtool utility: ethtool -S ethX
u32 match ip dport 5000 0xffff action skbedit queue_mapping 0
u32 match ip sport 5000 0xffff action skbedit queue_mapping 0
u32 match ip dport 5001 0xffff action skbedit queue_mapping 1
u32 match ip sport 5001 0xffff action skbedit queue_mapping 1
u32 match ip dport 5002 0xffff action skbedit queue_mapping 2
u32 match ip sport 5002 0xffff action skbedit queue_mapping 2
u32 match ip dport 5003 0xffff action skbedit queue_mapping 3
u32 match ip sport 5003 0xffff action skbedit queue_mapping 3
u32 match ip dport 5004 0xffff action skbedit queue_mapping 4
u32 match ip sport 5004 0xffff action skbedit queue_mapping 4
u32 match ip dport 5005 0xffff action skbedit queue_mapping 5
u32 match ip sport 5005 0xffff action skbedit queue_mapping 5
u32 match ip dport 5006 0xffff action skbedit queue_mapping 6
u32 match ip sport 5006 0xffff action skbedit queue_mapping 6
u32 match ip dport 5007 0xffff action skbedit queue_mapping 7
u32 match ip sport 5007 0xffff action skbedit queue_mapping 7
dcbtool is used to query and set the DCB settings of a DCB capable Ethernet interface. It connects to the client interface of lldpad to perform these operations. dcbtool will operate in interactive mode if it is executed without a command. In interactive mode, dcbtool also functions as an event listener and will print out events received from lldpad as they arrive.
SYNOPSIS dcbtool -h dcbtool -v dcbtool [-rR] dcbtool [-rR] [command] [command arguments]
OPTIONS -h shows the dcbtool usage message
-v shows dcbtool version information
-r displays the raw lldpad client interface messages as well as the
readable output.
-R displays only the raw lldpad client interface messagesCOMMANDS help shows the dcbtool usage message
ping test command. The lldpad daemon responds with "PONG" if the
client interface is operational.
license
displays dcbtool license information
quit exits from interactive mode
The following commands interact with the lldpad daemon to manage the dae-
mon and DCB features on DCB capable interfaces.
lldpad general configuration commands:
------------------------------------
<gc|go> dcbx
gets the configured or operational version of the DCB capabili-
ties exchange protocol. If different, the configured version
will take effect (and become the operational version) after lldpad
is restarted.
sc dcbx v:[1|2]
sets the version of the DCB capabilities exchange protocol which
will be used the next time lldpad is started. The default version
is 2. Changes take effect on next lldpad restart.
Information about version 1 can be found at:
<http://download.intel.com/technology/eedc/dcb_cep_spec.pdf>
Information about version 2 can be found at:
<http://www.ieee802.org/1/files/public/docs2008/az-wadekar-dcbx-
capability-exchange-discovery-protocol-1108-v1.01.pdf>
DCB per-interface commands:
---------------------------
gc <ifname> <feature>
gets configuration of feature on interface ifname.
go <ifname> <feature>
gets operational status of feature on interface ifname.
gp <ifname> <feature>
gets peer configuration of feature on interface ifname.
sc <ifname> <feature> <args>
sets the configuration of feature on interface ifname.
feature may be one of the following:
------------------------------------
dcb DCB state of the port
pg priority groups
pfc priority flow control
app:<subtype>
application specific data
ll:<subtype>
logical link status
subtype can be:
---------------
0|fcoe Fiber Channel over Ethernet (FCoE)
args can include:
-----------------
e:<0|1>
controls feature enable
a:<0|1>
controls whether the feature is advertised via DCBX to the peer
w:<0|1>
controls whether the feature is willing to change its opera-
tional configuration based on what is received from the peer
[feature specific args]
arguments specific to a DCB feature
Feature specific arguments for dcb:
-----------------------------------
on|off enable or disable DCB for the interface. The go and gp commands
are not needed for the dcb feature. Also, the enable, advertise
and willing parameters are not required.
Feature specific arguments for pg:
----------------------------------
pgid:xxxxxxxx
Priority group ID for the 8 priorities. From left to right
(priorities 0-7), x is the corresponding priority group ID
value, which can be 0-7 for priority groups with bandwidth allo-
cations or f (priority group ID 15) for the unrestricted prior-
ity group.
pgpct:x,x,x,x,x,x,x,x
Priority group percentage of link bandwidth. From left to right
(priority groups 0-7), x is the percentage of link bandwidth
allocated to the corresponding priority group. The total band-
width must equal 100%.
uppct:x,x,x,x,x,x,x,x
Priority percentage of priority group bandwidth. From left to
right (priorities 0-7), x is the percentage of priority group
bandwidth allocated to the corresponding priority. The sum of
percentages for priorities which belong to the same priority
group must total 100% (except for priority group 15).
strict:xxxxxxxx
Strict priority setting. From left to right (priorities 0-7), x
is 0 or 1. 1 indicates that the priority may utilize all of the
bandwidth allocated to its priority group.
up2tc:xxxxxxxx
Priority to traffic class mapping. From left to right (priori-
ties 0-7), x is the traffic class (0-7) to which the priority is
mapped. (This setting is ignored for Intel 82598-based
adapters.)
Feature specific arguments for pfc:
-----------------------------------
pfcup:xxxxxxxx
Enable/disable priority flow control. From left to right (pri-
orities 0-7), x is 0 or 1. 1 indicates that the corresponding
priority is configured to transmit priority pause.
Feature specific arguments for app:< subtype>:
----------------------------------------------
appcfg:xx
xx is a hexadecimal value representing an 8 bit bitmap where bits
set to 1 indicate the priority which frames for the applications
specified by subtype should use. The lowest order bit maps to
priority 0.
Feature specific arguments for ll:<subtype>:
--------------------------------------------
status:[0|1]
For testing purposes, the logical link status may be set to 0 or
1. This setting is not persisted in the configuration file.EXAMPLES Enable DCB on interface eth2
dcbtool sc eth2 dcb on
Assign priorities 0-3 to priority group 0, priorities 4-6 to priority
group 1 and priority 7 to the unrestricted priority. Also, allocate
25% of link bandwidth to priority group 0 and 75% to group 1.
dcbtool sc eth2 pg pgid:0000111f pgpct:25,75,0,0,0,0,0,0
Enable transmit of Priority Flow Control for priority 3 and assign FCoE
to priority 3.
dcbtool sc eth2 pfc pfcup:00010000
dcbtool sc eth2 app:0 appcfg:08How did Intel verify their DCB solution?
Answer - The Intel solution is continually evolving as the relevant standards become solidified and more vendors introduce DCB capable systems. That said, we initially used test automation to verify the DCB state machine. As the state machine became more robust and we had DCB capable hardware, we began to test back to back with our adapters. Finally, we introduced DCB capable switches in our test bed.
Prior to kernel 2.6.26, TSO will be disabled when the driver is put into DCB mode.
A TX unit hang may be observed when link strict priority is set when a large amount of traffic is transmitted on the link strict priority.
lldpad, lldptool and dcbtool - LLDP daemon with DCB support and command line utilities for LLDP and DCB configuration Copyright(c) 2007-2010 Intel Corporation.
Portions of lldpad, lldptool and dcbtool are based on: hostapd-0.5.7 Copyright (c) 2004-2007, Jouni Malinen j@w1.fi
This program is free software; you can redistribute it and/or modify it under the terms and conditions of the GNU General Public License, version 2, as published by the Free Software Foundation.
This program is distributed in the hope it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
The full GNU General Public License is included in this distribution in the file called "COPYING".
Contact Information: LLDP-devel Mailing List lldp-devel@open-lldp.org Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497