Permalink
7c01a27 Nov 24, 2015
@wowczarek @gvnn3 @skreuzer @j123b567
154 lines (108 sloc) 6.09 KB
PTP Daemon Version 2 Build and Installation Instructions
15 June 2015
George V. Neville-Neil, Wojciech Owczarek
The PTP Daemon is known to work on FreeBSD, NetBSD, OpenBSD,
OpenSolaris and its derivatives, Mac OS X and Linux systems.
*) Requirements
You will need the following tools and libraries to build ptpd2:
1. GNU autotools (automake, autoconf, libtool etc.) - essential
2. PCAP libraries (libpcap/bpf) and development headers
(optional but recommended) - required for Ethernet transport
and generally recommended for message performance
*NOTE: Libpcap seems to be broken on OpenSolaris / OmniOS,
builds and runs, but no data is being received.
3. SNMP libraries (optional) - will allow building with SNMP
support
*) Building
1. cd into the root of the tree (which is where the file you're
reading resides)
2. If building from a repository such as Subversion, CVS or GIT, run:
autoreconf -vi
3. If building from a distribution .tar.gz package, you may run autoreconf -vi,
but in most cases running ./configure should be enough - you must run
autoreconf if configure or result in errors.
4. ./configure
The final section of the configure script output will show
the results of all library dependency checks and feature checks:
features and dependencies can be controled using the options
described below. Running ./configure --help will show all available
options.
*NOTE* On QNX systems, the configure script may not accept the
'rm' command implementation available, so if it complains,
run ./configure with the following environment variable:
export ACCEPT_INFERIOR_RM_PROGRAM=yes
========= configure script options for ptpd ========================
* To disable the statistics code in order to lower computational
expense and RAM usage, configure the build with the following options:
./configure --disable-statistics
* To build a slave-only version (with GM logic disabled), build with:
./configure --enable-slave-only
This will build a version incapable of running as master.
* As of 2.3.1, the default timer implamentation used for periodic
actions uses POSIX timers if only the OS supports it. To disable
POSIX timer support and use the old interval timer implementation,
the following switch can be used:
./configure --disable-posix-timers
* As of 2.3.1, support was added for multiple unicast destinations
(both GMs and slaves) - with negotiation (signaling) and without.
The default maximum number of unicast destinations (also the
maximum number of slaves for a signaling GM) is set to 128. This
setting has an impact on the amount of RAM used by PTPd. The
unicast table capacity can be changed to a maximum of 2048 using:
./configure --with-max-unicast-destinations=[16...2048]
It is possible to change this limit to more than 2048, but realistically
due to message scheduling, timer signal usage and single-threaded
architecture, anywhere above 1000 ptpd will show high CPU utilisation.
Ptpd has been successfully tested as unicast Telecom (negotiation) master
at 1000 slaves and high message rates (32/sec) and provided adequate
performance. When running high numbers of slaves, it is recommended
to use libpcap (ptpengine:use_libpcap=y) to allow easier matching
of Sync and FollowUp, unless the NIC supports software transmit
timestamps. Without those and without libpcap, ptpd loops packets
back into its own socket, so the followUp destination cannot be
matched with the Sync destination, and a lookup table is used.
Hash collissions are low, but they may happen, in which case
ptpd will iterate over a table of last timestamps. This is only and issue
in extreme scenarios.
* With certain NIC drivers on Linux providing the SO_TIMESTAMPING
functionality with software transmit timestamps (SOF_TIMESTAMPING_
TX_SOFTWARE), unpredictable transmit timestamp failures can occur,
which may cause unstable operation and in some cases high CPU usage
(see https://sourceforge.net/p/ptpd/discussion/469208/thread/4aea6e47).
While fixes have been implemented and have shown success,
there is still some risk. To disable SO_TIMESTAMPING support on Linux
completely, regardless of OS support, use:
./configure --disable-so-timestamping
* To enable experimental options use:
./configure --enable-experimental-options
This enables:
- running SO_TIMESTAMPING without verifying if NIC driver
supports the given timestamping options
- running message intervals outside of PTP specification
*NOTE* On QNX systems, when experimental options are enabled,
a clock_gettime approximation using CPU clock counter and
attaching to IRQ0 is used, and this is also used to retrieve
packet RX and TX timestamps, ignoring PCAP timestamps and
socket options. This is recommended for best performance,
but requires more testing before becoming the default.
by default in QNX, clock is incremented in ticks that
can be as long as 10 milliseconds, and PCAP and socket
timestamps are only accurate to several milliseconds.
5. make
6. Read the manual pages ptpd2(8) and ptpd2.conf(5). The man pages
are the most complete source of configuration information.
ptpd2 itself also provides an extensive help:
run ptpd2 --help to see the short help
run ptpd2 --long-help to see the long help for all settings
run ptpd2 -e [key:setting] to display help for a single setting
7. Update test/client-e2e-socket.conf so that its
"ptpengine:interface = " setting points to a network interface on
your test machine that can see PTP packets from a grandmaster.
8. Test it in place: ./src/ptpd2 -c test/client-e2e-socket.conf
9. Check the log output of the daemon in /var/run/ptpd2.event.log
Check statistics output of the daemon in /var/run/ptpd2.stats.log
Check the status file /var/run/ptpd2.status.log
10. If the results look good, make install
The daemon may work on other POSIX based systems but this is not
guaranteed. Patches and fixes are welcome on the SourceForge page:
http://ptpd.sf.net