Hello,
Our experiment based on MIDAS 2.x DAQ.
I'm using several identical frontend-%d with only lam source & event id changed,
running on 2 computers(~3frontends per one).
Each recieve about 10k Events (Max_SIZE =8*1024, but usually it is less then
sizeof(DWORD)*400) per 7sec.
With no mlogger running it works just fine, but when I'm starting mlogger (on 3-d
computer with mserver running)... looking at ethernet stat graph first 2-3 spills
goes well, with one peak per 7 sec, then it becomes junky and everithing crushed
(mlogger and frontends).
I tried to increase SYSTEM buffer and restart everything. What I saw was Logger
writes only half of recieved events from sum of frontends, it stays running for
awhile ~15minutes. If I push STOP button before crashing, mlogger continious
writing data on disk enough priod of time.
I will try to look at disk usage for bad sectors @HDD, but may be there is an easy
way to fix this problem and i did something wrong.
structure of frontend has code like
EQ_POLLED , POLL for 500,
frontend_loop{
read big buffer with 10k events;bufferread=true;
}
poll_event{
for (i=0;i<count;i++){
if (bufferread) lam=1;
if (!test) return lam;
}
return 0;
}
read_trigger{
bk_init32();
//fill event with buffer until current word!=0xffffffff
if (currentposition+2 >buffer_size) bufferread=false
}
|
Help needed, please. Suggestions.
Thanks, Alexey. |