Problem in CosmoMC with mpirun

[srijita sinha]

Hi,

I am using CosmoMC in an SGE cluster with specifications:

Code: Select all

(CosmoPython) [narayan@dirac CosmoMC]$ lscpu 
Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                40
On-line CPU(s) list:   0-39
Thread(s) per core:    1
Core(s) per socket:    20
Socket(s):             2
NUMA node(s):          2
Vendor ID:             GenuineIntel
CPU family:            6
Model:                 85
Model name:            Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz
Stepping:              4
CPU MHz:               2400.000
BogoMIPS:              4800.00
Virtualization:        VT-x
L1d cache:             32K
L1i cache:             32K
L2 cache:              1024K
L3 cache:              28160K
NUMA node0 CPU(s):     0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38
NUMA node1 CPU(s):     1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39
Flags:                 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb cat_l3 cdp_l3 intel_pt tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts

The submit file looks like

Code: Select all

# The shell used to run the job
#$ -S /bin/bash
# The name of the parallel queue to submit the job to
#$ -N cosmomc
# Define the parallel runtime environment and number of nodes
# NB: number of nodes is one more than needed as one copy resideson the master node
#$ -pe mpi 8
# Use location that job was submitted as working directory
#$ -cwd
module load intel-compiler
source /home/softwares/intel/impi/2019.1.144/intel64/bin/mpivars.sh
mpirun -np 8 ./cosmomc test.ini\
#mpiexec -iface ib0 -f /home/softwares/Hostfiles/hosts.ifc -np $NSLOTS ./cosmomc test.ini

With the command

Code: Select all

mpirun -np 8 ./cosmomc test.ini

8 chains are running for 800 samples and then after MPI communicating, only chain 0 is working. Data is stored in test_1.dat only and log is written in test_1.log. The output screen/file reads as

Code: Select all

MPI startup(): I_MPI_F77 environment variable is not supported.
MPI startup(): To check the list of supported variables, use the impi_info utility or refer to https://software.intel.com/en-us/mpi-library/documentation/get-started.
....
....
....
Chain:7 drag accpt:  0.4059041     fast/slow   67.91409     slow:         745
 Chain:5 drag accpt:  0.4245283     fast/slow   67.21973     slow:         760
 Chain:1 drag accpt:  0.4069051     fast/slow   68.56148     slow:         732
 Chain:2 drag accpt:  0.4176334     fast/slow   63.66264     slow:         744
 Chain:4 drag accpt:  0.4031355     fast/slow   67.36502     slow:         789
 Chain:0 drag accpt:  0.4422604     fast/slow   67.64106     slow:         755
 Chain:6 drag accpt:  0.4386952     fast/slow   68.45614     slow:         798
 Chain:5 drag accpt:  0.4211663     fast/slow   67.33900     slow:         823
 Chain:3 drag accpt:  0.4235294     fast/slow   67.64450     slow:         782
 Chain:7 drag accpt:  0.4040404     fast/slow   67.89487     slow:         818
 Chain           2  MPI communicating
 Chain           8  MPI communicating
 Chain           6  MPI communicating
 Chain           3  MPI communicating
 Chain           5  MPI communicating
 Chain           7  MPI communicating
 Chain           4  MPI communicating
 Chain:0 drag accpt:  0.4352679     fast/slow   67.54447     slow:         832
 Chain:0 drag accpt:  0.4325438     fast/slow   67.68958     slow:         902
 Chain:0 drag accpt:  0.4302103     fast/slow   67.58334     slow:         972
 Chain:0 drag accpt:  0.4316547     fast/slow   67.59238     slow:        1023

Whereas with the command line

Code: Select all

mpiexec -iface ib0 -f /home/softwares/Hostfiles/hosts.ifc -np $NSLOTS ./cosmomc test.ini

all the chains are stopped after some

proxy_upstream_control_cb

and

HYDI_dmx_poll_wait_for_event

errors!

Even if I do not submit the job in a queue and run in a master node, these problems prevail.

Anyone, please help!

Srijita

[Terry Kovacs]

We are having exactly the same problem. Any ideas or hints as to how to debug would be helpful

Chain:2 drag accpt: 0.3939962 fast/slow 71.06418 slow: 483
Chain:6 drag accpt: 0.4216867 fast/slow 71.34849 slow: 462
Chain:1 drag accpt: 0.4418985 fast/slow 72.58334 slow: 552
Chain:3 drag accpt: 0.4411765 fast/slow 71.12830 slow: 569
Chain 8 MPI communicating
Chain 6 MPI communicating
Chain 4 MPI communicating
Chain 2 MPI communicating
Chain 5 MPI communicating
Chain 7 MPI communicating
Chain 3 MPI communicating
Chain:0 drag accpt: 0.4285714 fast/slow 71.35812 slow: 511
Chain:0 drag accpt: 0.4361874 fast/slow 71.25044 slow: 567
Chain:0 drag accpt: 0.4484305 fast/slow 71.13866 slow: 613

[Antony Lewis]

I am now also seeing this also with a recent intel Fortran (late 2018 or 2019) setup, maybe something changed. Did anyone find a solution?

[Antony Lewis]

The current advice seems to be simply not to use the 2019 Intel MPI, but some earlier version.