L2 Torture Tests of Using 6 Inputs Instead of 3
Conclusions from L2_Torture test
The cases have been separated in terms of # of large inputs on each PCI
bus.
The assumptions are that it does not matter which FILAR the inputs are
on if they are both on the
same PCI bus (i.e inputs in FILAR 1 and 2 have the same effect). This
has been observed numerous
times before.
Details of the test:
* Using 6 XFTs on 2 FILAR cards (0 and 1 for these tests)
* Using the merged old inputs to get reference for bunch counter etc
(location of merged input does
not change outcome of tests, size ~70 words)
* L1 accept rate: starts at ~1Hz and reaches 1.7kHz by about 600
events. Ran all tests for 1000
events.
4 Large inputs on 1 FILAR (1 PCI bus):
1) 4x(140-193 words)= total 560 - 773
- OK for ~1000 events
2) 3x(140-193) + 1x(280-386) = total 700 - 955
- 2 HRR in 1000 events: largest packet is corrupt
3) 4x(221-287) = total 884 - 1148
- Ran for only 50 events (<10Hz) then after each HRR only ~10 events
- Data corrupt on random packet
3 Large inputs on 1 FILAR (1 PCI bus):
1)3x(140-193) = total 420 - 579
- OK for ~1000 events
2) 2x(140-193) + 1x(280-386) = total 560 - 772
- OK for ~1000 events
3) 1x(140-193) + 2x(280-386) = total 700- 965
- OK for ~1000 events
4) 3x(280-386) = total 840 - 1158
- HRR after ~800 events (rate ~17.kHz)
- Data corrupt in last large packet to arrive to PC
- No more failured for ~800 events
5) 3x(221-286) = total 663 - 858
- OK for ~1000 events
6) 2x(221-286) + 1x(423-639) = total 865 - 1211
- HRR after ~800 events
- Data corruption on largest event + bunch counter mismatch
- Previous event for largest packet is where the problem lies: Shuld
have been ~530 words but was
837!!!
-> Looks like 2 events where merged together for previous event but
did not get corruption as
found trailer and header OK.
- Ran fine for next ~200 events
7) 1x(221-286)+2x(423-639) = total 1067-1564
- No HRR for ~1000 events
- Quite a few single events that were bad but never 3 in a row
8) 3x(423-639) = total 1269-1917 words
- Ran for ~300 events (rate <10Hz)
- then many HRR (approx every 50 events)
- Data corruption on large data size, bunch counter mismatch
- No problems seen in sizes of previous event
- Completely stopped running when rate went to ~1.7kHz (after ~630
events)
2 Large inputs on 1 FILAR:
1) 2x(560-760) = total 1120-1520
- No problems for ~1000 events (rate ~1.7kHz)
2) Did not try any higher....
Conclusions:
4 Large packets:
* Runs fine when max size on FILAR <770
* Few corrupted events when size ~950
* Breaks when max size ~1150
3 Large packets:
* Runs fine up to max size on FILAR < 960
* Only few HRR up to size ~1500
* Few HRR when rate ~10Hz-1.5kHz, dies for rate ~1.7kHz for max sizes
~1900
2 Large packets:
* No problems with total max size ~1500