Report: Latency and Loss of Timing Messages in the Timing System
Introduction
Starting in October 2019 the ECA-Tap module was added to the gateware of a few dedicated timing receivers. The ECA-Tap module provides the following statistics at the ECA input.
- # of messages received
- time differences: deadline - timestamp
- average (calculated from 'sum' and '# of messages')
- minimum
- maximum
- # of late messages
From this, diagnostic data can be derived. It is tried to connect timing receivers at different layers of network switches. As an exmample, comparing the statistics data from different network layers allows to extract data such as latency or loss of messages per layer.
PRO
For the location of the nodes in the production system see
here.
layer |
date |
node |
direction |
dt min [us] |
dt max [us] |
dt ave [us] |
av latency [us] |
total # of messages |
loss rate |
# of lates |
late rate |
1 |
2019-oct-11 |
DM |
TX |
891.0 |
998.7 |
~ 981.5 |
N/A |
-- |
N/A |
0 |
-- |
1 |
ident |
SCU diag |
RX |
879.1 |
990.1 |
941.6 |
39.9 |
5061751 |
N/A |
0 |
N/A |
2 |
ident |
SCU diag |
RX |
876.3 |
987.3 |
938.8 |
2.8 |
ident |
< 1.97e-7 |
0 |
< 1.97e-7 |
3 |
ident |
SCU diag |
RX |
873.8 |
984.8 |
936.2 |
2.6 |
ident |
< 1.97e-7 |
0 |
< 1.97e-7 |
1 |
2019-oct-15 |
DM |
TX |
891.0 |
998.7 |
~ 982.0 |
N/A |
-- |
N/A |
0 |
-- |
1 |
ident |
SCU diag |
RX |
790.1 |
990.2 |
942.2 |
40.0 |
19697158 |
N/A |
0 |
N/A |
2 |
ident |
SCU diag |
RX |
787.4 |
987.4 |
939.5 |
2.7 |
ident |
< 5.01e-8 |
0 |
< 5.01e-8 |
3 |
ident |
SCU diag |
RX |
784.7 |
984.8 |
936.9 |
2.6 |
ident |
< 5.01e-8 |
0 |
< 5.01e-8 |
1 |
2019-oct-21 |
DM |
TX |
891.0 |
998.7 |
~ 982.6 |
N/A |
-- |
N/A |
0 |
-- |
1 |
ident |
SCU diag |
RX |
790.1 |
990.2 |
942.2 |
40.4 |
36277751 |
N/A |
0 |
N/A |
2 |
ident |
SCU diag |
RX |
787.4 |
987.4 |
939.8 |
2.4 |
ident |
< 2.76e-8 |
0 |
< 2.76e-8 |
3 |
ident |
SCU diag |
RX |
784.7 |
984.8 |
937.2 |
2.6 |
ident |
< 2.76e-8 |
0 |
< 2.76e-8 |
1 |
2019-nov-14 |
DM (1) |
TX |
860.6 |
998.8 |
~ 987.4 |
N/A |
-- |
N/A |
0 |
-- |
1 |
ident |
SCU diag |
RX |
827.9 |
990.1 |
938.8 |
48.6 |
106706802 |
N/A |
1 |
N/A |
2 |
ident |
SCU diag |
RX |
825.0 |
988.3 |
936.0 |
2.8 |
ident |
< 9.37e-9 |
1 |
= 9.37e-9 |
3 |
ident |
SCU diag |
RX |
822.4 |
986,0 |
933.5 |
2.5 |
ident |
< 9.37e-9 |
1 |
= 9.37e-9 |
1 |
2019-nov-25 |
DM |
TX |
845.3 |
998.8 |
~ 983.0 |
N/A |
-- |
N/A |
0 |
-- |
1 |
ident |
SCU diag |
RX |
803.9 |
990.8 |
937.1 |
41.4 |
186760650 |
N/A |
4 |
N/A |
2 |
ident |
SCU diag |
RX |
801.5 |
988.3 |
934.4 |
2.4 |
ident |
< 4.48e-9 |
4 |
= 2.14e-8 |
3 |
ident |
SCU diag |
RX |
799.1 |
985.9 |
931.8 |
2.4 |
ident |
< 4.48e-9 |
4 |
= 2.14e-8 |
1 |
2019-dec-20 |
DM |
TX |
< 0 |
998.7 |
~ 980.0 |
N/A |
-- |
N/A |
0 |
-- |
1 |
ident |
SCU diag |
RX |
811.3 |
990.8 |
935.1 |
~40.0 |
265915522 |
N/A |
5 |
N/A |
2 |
ident |
SCU diag |
RX |
808.7 |
988.3 |
932.2 |
2.6 |
ident |
< 3.76e-9 |
5 |
= 1.88e-8 |
3 |
ident |
SCU diag |
RX |
806.0 |
985.9 |
929.8 |
2.7 |
ident |
< 3.76e-9 |
5 |
= 1.88e-8 |
Table: Typical numbers extracted from the production system. For each date, there are four rows of data. Description see text. If statistics has been reset, the DM (SCU) data are mareked by 2 (1).
The Data Master (DM) and three SCUs for diagnostic (SCU diag) are connected to a three layer WR network. The 'SCU diag' receive the messages sent by the DM. The above table shows the difference 'dt' (deadline - timestamp), the number of timing messages, and the number of late messages. The 'loss rate' is ratio of the number of lost messages to the total number of messages. The 'late rate' is the ratio of number of late messages to the total number of messages. The numbers for 'loss rate' and 'late rate' are always calculated to the row above. Please note:
- dt is measured at the input of the priority queue (DM)
- dt is measured at the input of the ECA ('SCU diag')
- after each measurement the counter of ECA-TAP are cleared
- 'total # of messages' is the sum of all measurements
UNILAC PRO
At LSB6 an experimental test setup is operated, see
here.
layer |
date |
node |
direction |
dt min [us] |
latency [us] |
# of messages |
loss rate |
# of lates |
late rate |
1 |
2019-oct-11 |
SCU PZ |
TX |
3051.4 |
N/A |
8e8 |
N/A |
0 |
< 1.25e-9 |
1 |
ident |
SCU diag |
RX |
3047.4 |
4.0 |
ident |
< 1.25e-9 |
0 |
< 1.25e-9 |
1 |
2019-oct-15 |
SCU PZ |
TX |
2992.6 |
N/A |
1.51e9 |
N/A |
0 |
< 6.62e-10 |
1 |
ident |
SCU diag |
RX |
2988.9 |
3.7 |
ident |
< 6.62e-10 |
0 |
< 6.62e-10 |
1 |
2019-oct-21 |
SCU PZ |
TX |
519.7 |
N/A |
2.66e9 |
N/A |
0 |
< 3.76e-10 |
1 |
ident |
SCU diag |
RX |
515.6 |
4.1 |
ident |
< 3.76e-10 |
0 |
< 3.76e-10 |
1 |
2019-oct-25 |
SCU PZ |
TX |
519.6 |
N/A |
3.34e9 |
N/A |
0 |
< 3.00e-10 |
1 |
ident |
SCU diag |
RX |
515.8 |
3.8 |
ident |
< 3.00e-10 |
0 |
< 3.00e-10 |
1 |
2019-nov-14 |
SCU PZ |
TX |
2831.2 |
N/A |
4.35e9 |
N/A |
0 |
< 2.30e-10 |
1 |
ident |
SCU diag |
RX |
2827.5 |
3.7 |
ident |
< 2.30e-10 |
0 |
< 2.30e-10 |
1 |
2019-nov-25 |
SCU PZ |
TX |
2499.7 |
N/A |
5.44e9 |
N/A |
0 |
< 1.84e-10 |
1 |
ident |
SCU diag |
RX |
2496.1 |
3.6 |
ident |
< 1.84e-10 |
0 |
< 1.84e-10 |
1 |
2019-dec-30 |
SCU PZ |
TX |
1792.4 |
N/A |
1.09e10 |
N/A |
0 |
< 9.18e-11 |
1 |
ident |
SCU diag |
RX |
1788.7 |
3.7 |
ident |
< 9.18e-11 |
0 |
< 9.18e-11 |
Table: Numbers extracted from UNILAC PRO. For each date, there are two rows of data. Description see text.
Two SCUs are connected to a layer1 WR switch. 'SCU PZ' serves as UNILAC Pulszentrale and sends timing messages to the WR network. 'SCU diag' receives the messages. The above table shows the difference 'dt' (deadline - timestamp), the number of timing messages, and the number of late messages. The 'loss rate' is ratio of the number of lost messages to the total number of messages. The 'late rate' is the ratio of number of late messages to the total number of messages. Please note:
- dt is measured at at the lm32 after the messages have been written to the EBM (SCU PZ)
- dt is measured at the input of the ECA ('SCU diag')
- after each measurement the counter of ECA-TAP are cleared
- 'total # of messages' is the sum of all measurements
--
DietrichBeck - 30 Dec 2019