Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: [OMPI users] exploitation of vampirTrace generated otf files.
From: Mathieu westphal (mathieu.westphal_at_[hidden])
Date: 2011-12-19 08:32:14


Hello

I'm generating otf file by compiling my programm with mpicc-vt.

It works very well and i can visualize the trace with the free Vampir.

But i need to analyze the results with some mathematical tools.

with vampir , i cannot extract any usefull data.

So i try to use otfdump to extract the data. It works well to, i get
lots of output like this :

(#12251) 6227944291 Leave: function 0, process 1, source 0
(#12252) 6228010545 Enter: function 136, process 1, source 0
(#12253) 6228010545 SendMessage: sender 1, receiver 3, group
1000000000, type 0, length 2880, source 0
(#12254) 6228012208 Leave: function 0, process 1, source 0
(#12255) 6228078341 Enter: function 126, process 1, source 0
(#12256) 6228727039 ReceiveMessage: receiver 1, sender 2, group
1000000000, type 0, length 4, source 0
(#12257) 6228727039 Leave: function 0, process 1, source 0
(#12258) 6228727787 Enter: function 136, process 1, source 0
(#12259) 6228727787 SendMessage: sender 1, receiver 3, group
1000000000, type 0, length 2880, source 0
(#12260) 6228729929 Leave: function 0, process 1, source 0
(#12261) 6228794962 Enter: function 136, process 1, source 0
(#12262) 6228794962 SendMessage: sender 1, receiver 3, group
1000000000, type 0, length 2880, source 0
(#12263) 6228796965 Leave: function 0, process 1, source 0
(#12264) 6228862900 Enter: function 136, process 1, source 0
(#12265) 6228862900 SendMessage: sender 1, receiver 3, group
1000000000, type 0, length 2880, source 0
(#12266) 6228864609 Leave: function 0, process 1, source 0
(#12267) 6228931011 Enter: function 136, process 1, source 0
(#12268) 6228931011 SendMessage: sender 1, receiver 3, group
1000000000, type 0, length 2880, source 0
(#12269) 6228932640 Leave: function 0, process 1, source 0
(#12270) 6228999027 Enter: function 136, process 1, source 0
(#12271) 6228999027 SendMessage: sender 1, receiver 3, group
1000000000, type 0, length 2880, source 0

What really interest me is in these lines :

(#12255) 6228078341 Enter: function 126, process 1, source 0
(#12256) 6228727039 ReceiveMessage: receiver 1, sender 2, group
1000000000, type 0, length 4, source 0
(#12257) 6228727039 Leave: function 0, process 1, source 0

and it's the time delta between the call to mpiRecv ( ReceiveMessage)
and the leave of mpiRecv, so in this case :
6228727039 - 6228078341 = 648698

But, what the unit of this value? why it dont correspond to anything in
Vampir?

After some test i empirically admit that

otfdumptime / 2267,999980709 = vampirTime (in µs)

I can work from that, but if anyone has any clue about that....

My guess is about some weird processor ticking count, but how vampir can
manage the conversion ?

Thanks for help