[Pw_forum] Could you please help me to cope with the error message

vega lew vegalew at hotmail.com
Tue Nov 18 17:51:52 CET 2008


Dear sir,
 
thank you so much for your suggestions.

 
> > I am suffering from the error message like this,> > [node1][0,1,12][btl_tcp_frag.c:202:mca_btl_tcp_frag_recv] > > mca_btl_tcp_frag_recv: readv failed with errno=104> > forrtl: error (78): process killed (SIGTERM)> > as explained many times: unless you have evidence of the opposite,> this kind of obscure errors is likely due to buggy compiler or buggy> libraries or even flakey hardware.
 
My Fortran / C compilers was intel 10.1.015 version. And MKL was 10.0.1.014. I could relax 72 atoms(openmpi) and 120 atoms(MPICH2) successfully using the compilers and MKL. If this version of compilers and MKL was not suitable for my hardware, could you please tell me which compiler and MKL was appropriate for my computers. Which compiler would be best choice for Q-E? 
By the way, my clusters have 8 computers composed of intel Q6600 cpu(quadcore), intel S3000AH mother boards, and 8G(2Gx4) physical memory per node (so 64G in total for the whole cluster).  The OS was redhat enterprise linux 4 update 4 for x86_64bit, therefore, the em64t verison of compiler and mkl was installed on the system.
 
> this might indicate a problem with excessive memory requirements
I tested the tasks using openmpi and MPICH2 respectively. QE with MPICH2 could relax 120 atoms, but QE with openmpi could relax 72 atoms. If the memory was not sufficient , why the relaxation of 84 atoms crashed using QE with openmpi but not the case for QE with MPICH2. And why QE with MPICH2 could calculate nearly the twice than the QE with openmpi?
 
My system to be calculated was composed of 48 Ti atoms 96 O atoms with a slab model. the ultrasoft pseudo(PW91) was downloaded for the pwscf website directly. The ecutwfc and ecutrho was 50 Ry and 500 Ry respectively.
 
Do you think 64G memory was not enough for such a task?
 
thank you for reading and kindly advised. I really appreciate your replying.
 
vega
 
Vega Lew (weijia liu)PH.D Candidate in Chemical EngineeringState Key Laboratory of Materials-oriented Chemical EngineeringCollege of Chemistry and Chemical EngineeringNanjing University of Technology, 210009, Nanjing, Jiangsu, China> Date: Fri, 14 Nov 2008 16:57:56 +0100> From: giannozz at democritos.it> To: pw_forum at pwscf.org> Subject: Re: [Pw_forum] Could you please help me to cope with the error message> > vega wrote:> > > I am suffering from the error message like this,> > [node1][0,1,12][btl_tcp_frag.c:202:mca_btl_tcp_frag_recv] > > mca_btl_tcp_frag_recv: readv failed with errno=104> > forrtl: error (78): process killed (SIGTERM)> > as explained many times: unless you have evidence of the opposite,> this kind of obscure errors is likely due to buggy compiler or buggy> libraries or even flakey hardware.> > > I could relax 72 atoms successfully with my system using openmpi. But > > when I wanted to relax 84 atoms, the error message stoped my > > calculation. Then I tried the mpich2 using the same system. With the > > help of mpich2 I could relax 120 atoms instead. But the error message > > bothered me again when I wanted to relax 132 atoms.> > this might indicate a problem with excessive memory requirements> > Paolo> -- > Paolo Giannozzi, Democritos and University of Udine, Italy> _______________________________________________> Pw_forum mailing list> Pw_forum at pwscf.org> http://www.democritos.it/mailman/listinfo/pw_forum
_________________________________________________________________
News, entertainment and everything you care about at Live.com. Get it now!
http://www.live.com/getstarted.aspx
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20081119/cbd562e9/attachment.html>


More information about the users mailing list