Link Layer Discovery Protocol (LLDP) Agent with
Data Center Bridging (DCB) for Intel(R) Network Connections
===========================================================
April 7, 2010
Contents
========
- Background
- Requirements
- Functionality
- How To Build a DCB capable system
- Setup
- Operation
- Testing
- dcbtool Overview
- FAQ
- Known Issues
- License
- Support
Background on DCB support
=========================
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.
Requirements
============
- Linux kernel version 2.6.29 or later.
- 2.6.29 or newer version of the "iproute2" package should be downloaded and
 installed in order to obtain a multi-queue aware version of the 'tc' utility.
 Check for new versions at:
 http://www.linuxfoundation.org/en/Net:Iproute2
- Version 2.5.33 of Flex should be installed (to support iproute2).
 Flex source can be obtained from http://flex.sourceforge.net/
- An up to date netlink library needs to be installed in order to compile
 lldpad.
- Version 1.3.2 or greater of libconfig must be installed.
 libconfig source can be obtained from http://www.hyperrealm.com/libconfig/
- For DCB features - a driver which supports the dcbnl rtnetlink interface
 in the kernel (e.g. the ixgbe driver for 82598-based or 82599-based Intel
 adapters).
Functionality
=============
lldpad 
 - Executes the Link Layer Discovery Protocol (LLDP) over all supported
 interfaces.
 - Executes the DCB capabilities exchange protocol (DCBX) to exchange DCB
 configuration with the peer device using LLDP on support interfaces.
 - Supports the versions of the DCB capabilities exchange protocol
 described here:
 - version 1
 <http://download.intel.com/technology/eedc/dcb_cep_spec.pdf>
 and here:
 - version 2
 <http://www.ieee802.org/1/files/public/docs2008/az-wadekar-dcbx-
 capability-exchange-discovery-protocol-1108-v1.01.pdf>
 - Retrieves and stores LLDP and DCB configuration to a configuration file.
 - Controls the DCB settings of the network driver based on the
 operation of the DCB capabilities exchange protocol. Interaction with
 a supporting network driver is achieved via DCB operations added to the
 rtnetlink interface in kernel 2.6.29.
 - Supports the following DCB features: Priority Group,
 Priority Flow Control, FCoE, and FCoE Logical Link Status.
 - Provides an interface for client applications to query and configure
 DCB features. Generates client interface events when the operational
 configuration or state of a feature changes.
lldptool
 - 
dcbtool
 - Interacts with lldpad via the client interface.
 - Queries the state of the local, operational and peer configuration for
 the supported DCB features.
 - Supports configuring the supported DCB features.
 - Interactive mode allows multiple commands to be entered interactively,
 as well as displaying event messages.
 - Enables or disables DCB for an interface.
How To Build a DCB-Capable System
=================================
Linux kernel install:
---------------------
1. Requires 2.6.29 kernel.
2. Untar and make the kernel. Listed below are the required kernel options:
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 
 
3. Build the kernel. 
4. 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
lldpad Application Install
--------------------------
1. 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.
	
2. 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
Options
-------
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
SETUP:
======
1. Load the ixgbe module. 
2. 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.
3. Enable DCB on the selected ixgbe port:
 dcbtool sc ethX dcb on
 
4. 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. 
DCBX Operation
==============
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.
1. qdisc and filter method
The skbedit action mechanism can be used in a tc filter to classify traffic
patterns to a specific queue_mapping value from 0-7. The ixgbe driver will
place traffic with a given queue_mapping value onto the corresponding hardware
queue and tag the outgoing frames with the corresponding 802.1p priority value.
Set up the multi-queue qdisc for the selected interface:
# tc qdisc add dev ethX root handle 1: multiq
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.
# tc filter add dev ethX protocol ip parent 1: u32 match ip dport 3260 \
0xffff action skbedit queue_mapping 4
# tc filter add dev ethX protocol ip parent 1: u32 match ip sport 3260 \
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.
# tc filter add dev ethX protocol 802_3 parent 1: handle 0xfc0e basic match \
'cmp(u16 at 12 layer 1 mask 0xffff eq 35078)' action skbedit queue_mapping 3
2. VLAN egress map method
Alternatively, traffic can be classified by configuring the egress map of a
VLAN interface. For example, if iSCSI traffic is desired to be on priority 4
and VLAN 101 is dedicated for iSCSI, then the following configuration could
be performed to map the skb_priorities on VLAN 101 to the 802.1p priority 4.
# vconfig set_egress_map ethX.101 0 4
# vconfig set_egress_map ethX.101 1 4
# vconfig set_egress_map ethX.101 2 4
...
# vconfig set_egress_map ethX.101 7 4
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.
Testing
=======
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
# tc qdisc add dev ethX root handle 1: multiq
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip dport 5000 0xffff action skbedit queue_mapping 0
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip sport 5000 0xffff action skbedit queue_mapping 0
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip dport 5001 0xffff action skbedit queue_mapping 1
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip sport 5001 0xffff action skbedit queue_mapping 1
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip dport 5002 0xffff action skbedit queue_mapping 2
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip sport 5002 0xffff action skbedit queue_mapping 2
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip dport 5003 0xffff action skbedit queue_mapping 3
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip sport 5003 0xffff action skbedit queue_mapping 3
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip dport 5004 0xffff action skbedit queue_mapping 4
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip sport 5004 0xffff action skbedit queue_mapping 4
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip dport 5005 0xffff action skbedit queue_mapping 5
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip sport 5005 0xffff action skbedit queue_mapping 5
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip dport 5006 0xffff action skbedit queue_mapping 6
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip sport 5006 0xffff action skbedit queue_mapping 6
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip dport 5007 0xffff action skbedit queue_mapping 7
# tc filter add dev ethX protocol ip parent 1: \
u32 match ip sport 5007 0xffff action skbedit queue_mapping 7 
dcbtool Overview
=================
 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 messages
 COMMANDS
 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:08
FAQ
===
- How 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.
Known Issues
============
- 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.
License
=======
 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".
Support
=======
 Contact Information:
 Issue Tracker: https://github.com/intel/openlldp/issues
 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497