Pelegant is taking too much time

Moderators: cyao, michael_borland

Post Reply
arahim
Posts: 26
Joined: 08 Jun 2012, 05:28

Pelegant is taking too much time

Post by arahim » 01 May 2019, 09:13

I am running run2.ele on sequential and parallel version of elegant. On seq. elegant run is completed in ~ 5 minutes, while with pelegant program is still running even half an hour is passed. I am using pelegant with 12 processors. Will the syntax of the input file will be same in elegant and pelegant or some thing I have to change.
Attachments
test_single_bunch.out
(508.91 KiB) Downloaded 305 times
test_lattice.lte
(1.89 KiB) Downloaded 313 times
run2.ele
(315 Bytes) Downloaded 313 times

michael_borland
Posts: 1927
Joined: 19 May 2008, 09:33
Location: Argonne National Laboratory
Contact:

Re: Pelegant is taking too much time

Post by michael_borland » 02 May 2019, 13:39

The problem is that you are asking for output of the particle coordinates every turn. Depending on your file system and the amount of computational work, this can have a major impact on overall performance. If you ask for every 10th turn, e.g.,

Code: Select all

w1: watch,filename="%s.w1",mode="coord",interval=10
or only the first 1000 particles, e.g.,

Code: Select all

w1: watch,filename="%s.w1",mode="coord",interval=1,start_pid=1,end_pid=1000
you'll get better performance.

You may also get some improvement from optimizing the write buffer size for MPI IO, e.g.,

Code: Select all

&global_settings
        mpi_io_write_buffer_size = 1000000
&end
--Michael

arahim
Posts: 26
Joined: 08 Jun 2012, 05:28

Re: Pelegant is taking too much time

Post by arahim » 23 Oct 2020, 00:42

How should we select the optimum number of nodes and and optimium number of processors to track the single bunch in a storage ring to study the single bunch instabilities. I have taken number of particles 100000 in a bunch and tracking 500000 turns.

Post Reply