A computer in the control room then gets all the TAI-
stamped acquisition data and displays it coherently on a
single screen. The operators then see this whole distributed
system as a simple oscilloscope to which all their signals
are connected. This would of course be impossible with
a real oscilloscope, but it can be done with WR’s phase-
compensated distribution of TAI, appropriate WR-enabled
nodes and software.
THE WR COMMUNITY
The WR project is a distributed endeavour whose collab-
orative model is heavily inspired by that typically used in
Free/Open Source software projects. All the intermediate
and final results in the project are made available through
open source licences. We use mostly the GNU General
Public License (GPL) and the GNU Lesser General Public
License (LGPL) for licensing software, LGPL for gateware
(HDL files) and the CERN Open Hardware Licence [8] for
hardware designs.
WR users are very often developers and vice versa. The
ever-growing list of interested institutes and companies [9]
has enlarged the scope of the project well beyond the origi-
nal domain of particle accelerators. At CERN, the effort of
dissemination for this technology has been coordinated in
collaboration with the Knowledge Transfer Group.
Thanks to its open nature, the project has benefited from
extensive contributions and ideas of potential and con-
firmed users. These exchanges typically happen on the
project mailing list, which is web-archived and integrated
with the project website [1]. Peer review over email or in
dedicated workshops establishes a meritocracy in which the
best ideas move forward. Openness is also key in avoiding
potential vendor lock-in situations and welcoming com-
mercial partners as any other contributor.
PROJECT STATUS AND OUTLOOK
At the time of this writing (August 2013) the WR switch
is a mature commercially-supported hardware product with
a stable release for its gateware and software. WR nodes
have been designed and validated in a variety of formats,
and they have consistently shown sub-ns synchronisation
accuracy. This section provides a quick overview of some
of the most important upcoming efforts in the project.
Synchronisation Performance
The results in Fig. 8 show that the original goals for WR
in terms of synchronisation have been achieved. In some
applications though, it is important to improve accuracy as
much as possible. While the WR delay model [10] and
compensation mechanism cover changes in fibre temper-
ature appropriately, they do not compensate for the delay
fluctuations induced by temperature variations in the nodes
and switches themselves. We believe these effects can ac-
count for tens of picoseconds or even more in particularly
adverse circumstances.
Tests in a climatic chamber [11] have shown that the de-
pendency of delay with respect to temperature is mostly
linear, so an evolution of the WR delay compensation algo-
rithm which takes into account on-board temperature mea-
surements should help in this respect. The modified algo-
rithm could include a linear model of the dependency or a
table of delay vs. temperature values resulting from offline
calibration.
Switch Evolution
The efforts on switch gateware and software will go in
the direction of providing better support for remote man-
agement, diagnostics and robust data transmission in a WR
network.
A Port Statistics block (Pstats, see Fig. 7) will be added
to provide counts of transmitted, received and dropped
frames per port, along with other counts relevant for as-
certaining the state of health of the network. A Time-
Aware Traffic Shaper Unit (TATSU) will allow blocking
selected output queues for some time so that high-priority
frames can go through the switch without colliding with
low-priority frames being sent at that moment through a
port. This will avoid having to wait until the end of trans-
mission of those frames before taking ownership of the
port.
In addition, a Topology Resolution Unit (TRU) will
add hardware support to the process of providing redun-
dant loop-free topologies for a network. This will result
in a faster topology switch-over compared to traditional
software-based methods such as the Rapid Spanning Tree
Protocol (RSTP). Figure 12 illustrates why fast switch-over
can be important in a WR network.
Figure 12: Forward Error Correction used to correct for
frame loss.
In this example, a WR node decomposes a control mes-
sage into four Ethernet frames which have been encoded
using Forward Error Correction (FEC), in such a way that
receiving any two of those frames allows the receiving node
to re-constitute the original control message. The use of
this type of FEC scheme is well adapted to networks in
which a late frame is a wrong frame, such as the one used
for accelerator timing at CERN. This precludes the use of
protocols which require re-trials in case of transmission er-
rors, such as the Transmission Control Protocol (TCP).
If the switch can enable a redundant port very quickly
after detecting a fatal condition in another port, and if the
switch-over time is lower than that corresponding to the
transmission of one of the frames in the figure, the receiv-
ing node will be able to re-constitute the message with the
remaining frames. Tests at CERN have shown that this kind
Proceedings of IBIC2013, Oxford, UK THBL2
Time Resolved Diagnostics and Synchronization
ISBN 978-3-95450-127-4
941
Copyright
c
○ 2013 by JACoW — cc Creative Commons Attribution 3.0 (CC-BY-3.0)