-
Notifications
You must be signed in to change notification settings - Fork 35
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Malloc problem #71
Comments
Hi,
|
Hi, lsmem output:
For what concern the anajob I think that you want to see the "size" on the input, I'm reconstructing a single signal event:
with 1000 bkg events like this ones:
For the last bullet I need to wait ten hours, the time required by the reconstruction job, Thanks for you help. ale |
Hi, overlaying 1000 times this amount of background is quite big. I guess you are reaching the limitations due to the OS on memory allocation by In particular:
If you know which event it is in particular, you can try to start your job at this event and before writing the event with the |
Another option might be to drop the MCParticle collection of the overlay events - this is what we do for the e+e- pair background overlay in ILC... |
This is the plot of the rss, the machine has 32Gb of RAM + 4Gb of swap. @rete I have also the suspect that there is a limitations due to the OS on memory allocation @gaede I'm trying your option The problem is that this is only a test, in production I'll have 1 signal event plus about 15k bkg events like the ones that I have posted, so probably I need a "more robust" workaround. thanks |
Is this still a problem? |
I guess it will give the same error. The implementation is similar in the new SIO. |
But as far as I can tell malloc is no longer called with an unsigned int (maximum 4 GB though) The documentation you link above is only for a particular implementation, in principle allocation should work for larger blocks But unsigned int was used before Line 483 in 05e5bb1
Line 130 in 05e5bb1
So I hope we don't have to guess. |
Unfortunately this seems to be not enough to fix the problem |
Hi,
I'm using ILCSoft for muoncollider simulation/reconstruction. At the end of a long reconstruction
I have this message (activating the VERBOSE flag in lcio):
A runtime error occured - (uncaught exception):
lcio::IOException: [SIOWriter::writeEvent] couldn't write event record to stream: Output_REC_000_slcio0
Marlin will have to be terminated, sorry.
And in the log file:
[ VERBOSE "Output_REC"] SIO: [Output_REC_000_slcio0/LCEvent/] Allocated a 536870912(0x20000000) byte buffer
[ VERBOSE "Output_REC"] SIO: [Output_REC_000_slcio0/LCEvent/] Allocated a 1073741824(0x40000000) byte buffer
[ VERBOSE "Output_REC"] SIO: [Output_REC_000_slcio0/LCEvent/] Buffer allocation failed
So it seems that it is not able to allocate 2Gb of buffer.
I have tried to play with the ulimit command:
max memory size (kbytes, -m) unlimited
stack size (kbytes, -s) 2202000
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited
But it doesn't help.
Do you have any suggestions?
The text was updated successfully, but these errors were encountered: