Specifications

CHAPTER 2. NETWORK PERFORMANCE MEASUREMENTS SYSTEMS 10
The computers involved in the system run a Linux with the nanokernel patches. This
allows a nanosec ond resolution of the system clock. The PPS signal is distributed
from the GPS to the serial ports of all computers via standard Cat5 cabling. The
PPS distribution contains also a voltage level converter from TTL to RS-232. The
precision obtained with this setup is of 10 microseconds.
The time synchronization system for K2K experiment
The K2K is a physics experiment developed at the Japanese national high energy
physics laboratory [10]. The experiment needs synchronized clocks in two sites that
are at a distance of 250km one from the other. The solution involved GPS receivers
at each site and some custom made local clock boards (LTC). The clock boards have
a 50MHz free-running counter. The GPS has two outputs: the PPS signal whose
leading edge is correlated with the UTC second, and an Ascii stream with the full
GPS data (time and position). The PPS signal is used to calibrate the clock board
(to see the average number of clock ticks per UTC second).
Upon receipt of each event that needs a time stamp, the LTC count and the GPS
Ascii data are latched and recorded. The LTC count provides the fractional second
of the time, down to 20 nsec precision, accurately synchronized with UTC within
100 nsec. The Ascii data provides the date and coarse time down to seconds. The
accuracy that is obtained is of the order of hundreds of nanosec onds.
2.3 Some observations
Most commercial testers that were presented are dedicated devices that offer a set of services
but that can’t be easily modified by the user without the help of the manufacturer. We
are interested in a system that is powerful enough but still gives full access to its internal
functioning. All the systems described above provide similar functionality for network
testing.
For accurate long distance measurements the solution is the GPS for time synchronization.
If one wants to build a testing system using off-the-shelf PCs and network cards the solution
for accurate timing is the NTP. This protocol can transfer time from host workstations con-
figured as a time server or from off-the-shelf time server products. These network transfer
techniques offer performance in the 1-10 millisecond synchronization accuracy range. This
performance can be extended further, to the 1- 10 microsecond range, by employing direct
connection time transfer techniques: time code, serial time messaging, and 1 PPS/reference
frequency signals. The time resolution on the host computers can be also improved up to
the nanosecond le vel by using some internal registers found in modern processors. The
accuracy obtained is of the order of tens of microseconds.
The most accurate method is the one that uses directly a GPS reference and special clock
cards (as in the K2K experiment) this is very similar to the method that will be described
in this report.
In the following we shall describe the architecture of the testing system we shall see
that the most important parts use customized hardware. This is because a pure software
solution cannot handle the performance measurements at full Gigabit line speeds.