ARMCI problem on SGI Altix


Clicked A Few Times
Dear NWChem users,

I've had the problems with NWChem which I describe here:

http://tiger.chem.uw.edu.pl/staff/lrajchel/nwchem_post.txt

(I cannot describe them here as I keep getting the "The specified URL cannot be found" error on the webpage on several browsers, maybe that's attributed to the NWChem web page settings. The funny thing is that it suffices to write c.mpile word in starting the new thread to get that error (replace . with o, of course). But that's another story.)

Sorry about this strange workaround to post my thread and thanks for your replies.

Forum Vet
You specify ARENA as the ARMCI_NETWORK? I don't believe that network exists as a choice. This might be your root cause...

Bert


Quote:Luke Mar 21st 6:13 pm
Dear NWChem users,

I've had the problems with NWChem which I describe here:

http://tiger.chem.uw.edu.pl/staff/lrajchel/nwchem_post.txt

(I cannot describe them here as I keep getting the "The specified URL cannot be found" error on the webpage on several browsers, maybe that's attributed to the NWChem web page settings. The funny thing is that it suffices to write c.mpile word in starting the new thread to get that error (replace . with o, of course). But that's another story.)

Sorry about this strange workaround to post my thread and thanks for your replies.

Clicked A Few Times
Quote:Bert Mar 29th 8:54 pm
You specify ARENA as the ARMCI_NETWORK? I don't believe that network exists as a choice. This might be your root cause...


Dear Bert,

I tried using other options for ARMCI_NETWORK or simply leaving the variable blank - none solution worked.

Forum Vet
Lets go step by step here:

Tell me about your cluster. What are the processors, what is the interconnect?

Bert


Quote:Luke Mar 31st 12:08 am
Quote:Bert Mar 29th 8:54 pm
You specify ARENA as the ARMCI_NETWORK? I don't believe that network exists as a choice. This might be your root cause...


Dear Bert,

I tried using other options for ARMCI_NETWORK or simply leaving the variable blank - none solution worked.

Clicked A Few Times
Quote:Bert Apr 1st 6:11 pm
Lets go step by step here:

Tell me about your cluster. What are the processors, what is the interconnect?

Bert


It's the SGI Altix 350 Cluster with 16 Itanium processors

cat /proc/cpuinfo
...
processor  : 15
vendor     : GenuineIntel
arch       : IA-64
family     : Itanium 2
model      : 1
revision   : 5
archrev    : 0
features   : branchlong
cpu number : 0
cpu regs   : 4
cpu MHz    : 1500.000000
itc MHz    : 1500.000000
BogoMIPS   : 2239.75


The interconnect is NUMAlink, according to the SGI website:

Quote:

High-performance, shared-memory SGI® NUMAlink® interconnect for unparalleled I/O at 6.4GB/second


Some more info:

uname -a
Linux (...) 2.4.21-sgi305rp05041818_10155 #1 SMP Mon Apr 18 19:04:07 PDT 2005 ia64 ia64 ia64 GNU/Linux

Forum Vet
Okay,

You should try the following:

1. make realclean

2. Unset ARMCI_NETWORK

3. make NWChem

Can you log your compilation and send it to me if it fails again.

Bert


Quote:Luke Apr 7th 10:34 pm
Quote:Bert Apr 1st 6:11 pm
Lets go step by step here:

Tell me about your cluster. What are the processors, what is the interconnect?

Bert


It's the SGI Altix 350 Cluster with 16 Itanium processors

cat /proc/cpuinfo
...
processor  : 15
vendor     : GenuineIntel
arch       : IA-64
family     : Itanium 2
model      : 1
revision   : 5
archrev    : 0
features   : branchlong
cpu number : 0
cpu regs   : 4
cpu MHz    : 1500.000000
itc MHz    : 1500.000000
BogoMIPS   : 2239.75


The interconnect is NUMAlink, according to the SGI website:

Quote:

High-performance, shared-memory SGI® NUMAlink® interconnect for unparalleled I/O at 6.4GB/second


Some more info:

uname -a
Linux (...) 2.4.21-sgi305rp05041818_10155 #1 SMP Mon Apr 18 19:04:07 PDT 2005 ia64 ia64 ia64 GNU/Linux

Clicked A Few Times
Quote:Bert Apr 14th 5:57 pm
Okay,

You should try the following:

1. make realclean

2. Unset ARMCI_NETWORK

3. make NWChem

Can you log your compilation and send it to me if it fails again.


Hi Bert,

I cleared the ARMCI_NETWORK variable, cleared the NWChem tree and recompiled - it finished with the same error as before, and the complete compilation log is here.

Thanks for the help so far!

Forum Vet
That's weird, this should not even have been set with Itanium processors.

Can you do a "uname -m" on the node. It should say ia64.

Another way to eliminate is by unsetting USE_MPI, in tools do a make clean and make CC=icc FC=ifort, and a make link FC=ifort in src.

Bert


Quote:Luke Apr 19th 8:22 pm
Quote:Bert Apr 14th 5:57 pm
Okay,

You should try the following:

1. make realclean

2. Unset ARMCI_NETWORK

3. make NWChem

Can you log your compilation and send it to me if it fails again.


Hi Bert,

I cleared the ARMCI_NETWORK variable, cleared the NWChem tree and recompiled - it finished with the same error as before, and the complete compilation log is here.

Thanks for the help so far!

Clicked A Few Times
Quote:Bert Apr 19th 11:35 pm
That's weird, this should not even have been set with Itanium processors.

Can you do a "uname -m" on the node. It should say ia64.

Another way to eliminate is by unsetting USE_MPI, in tools do a make clean and make CC=icc FC=ifort, and a make link FC=ifort in src.

Bert


Dear Bert,

I've successfully managed to c*mpile [again problems with the forum...] NWChem on SGI ia64 and the detailed description is here:
http://tiger.chem.uw.edu.pl/staff/lrajchel/nwchem-6.0-install.txt

Thanks for your help!


Forum >> NWChem's corner >> Compiling NWChem