TAPRIO on Zenoh experimental result
p=1, 2, 3
on pub
sched-entry S 03 1000000 \
sched-entry S 05 3000000 \
sched-entry S 09 5000000 \
p1: 27438.11878
p2: 28825.8863
p3: 28820.70199
on sub
sched-entry S 03 1000000 \
sched-entry S 05 3000000 \
sched-entry S 09 5000000 \
p1: 44293.78231
p2: 46067.4276
p3: 41663.88396
delete TAPRIO settings on both pub and sub
p1: 79584.90832
p2: 74981.0241
p3: 74930.43537
p=1, 2,3
sched-entry S 03 1000000 \
None of them received, replacing 03 with 05 or 09 also didn’t work. Adding one more 0 to 1000000 didn’t work either.
sched-entry S 03 1000000 \
sched-entry S 09 5000000 \
None of them received.
sched-entry S 05 3000000 \
None of them received.
### on pub
sched-entry S 03 1000000 \
sched-entry S 05 3000000 \
sched-entry S 09 5000000 \
All three received, but without distinction.
p=1
on pub
sched-entry S 03 1000000 sched-entry S 05 3000000 sched-entry S 09 5000000
p1: 29727.14288
p=2
on pub
sched-entry S 03 1000000 sched-entry S 05 3000000 sched-entry S 09 5000000
p2: 94691.03561
p=3
on pub
sched-entry S 03 1000000 sched-entry S 05 3000000 sched-entry S 09 5000000
p3: 158004.1489
p=1 2
on pub
sched-entry S 03 1000000 sched-entry S 05 3000000 sched-entry S 09 5000000
p1: 36350.44085
p2: 36450.19939
Numbers are fluctuating, not stable.
p=1 1
Mainly testing if different threads receiving the same packets would yield different results. Opened two threads subscribing to the same topic.
Adjacent numbers are almost the same. It can be inferred that when there is a large difference between numbers received with different priorities, they are not from the same batch.
pub_p=1 2 3, sub_p=2
on pub
sched-entry S 03 1000000 sched-entry S 05 3000000 sched-entry S 09 5000000
p2: 94532.33984
Although the performance when there are three priorities on the publisher side and only one on the subscriber side is the same as when only one priority is sent, it is reasonable because when there is no subscriber to receive, Zenoh will stop sending packets for that session. We believe the problem is more likely to occur on the publisher side.