Hi, At the page 97 of the siesta-3.0-beta manual,it indeed says :"The present version of the code is only parallelized over k-points, so the number of nodes should not be bigger than the number of k-points."
I thought it was the character of the tool tbtrans! But when I run transiesta setting Diag.ParallelOverK false,it always comes the error : ----------------- Begin LEFT unit cell: 21.7981 0.0000 0.0000 0.0000 18.8777 0.0000 0.0000 0.0000 13.3485 p1_28306: (5434.726562) net_recv failed for fd = 3 p1_28306: p4_error: net_recv read, errno = : 104 rm_l_1_28331: (5435.109375) net_send: could not write to fd=5, errno = 32 bm_list_28301: (5436.523438) wakeup_slave: unable to interrupt slave 0 pid 28300 /home/zgp/software/mpich-1.2.7/bin/mpirun.ch_p4: line 243: 28300 Segmentation fault /home/zgp/calcu/transiesta/saifen-saizuo/sr-2/transiesta -p4pg /home/zgp/calcu/transiesta/saifen-saizuo/sr-2/PI28194 -p4wd /home/zgp/calcu/transiesta/saifen-saizuo/sr-2 p2_28336: (5434.589844) net_recv failed for fd = 3 p2_28336: p4_error: net_recv read, errno = : 104 rm_l_2_28361: (5434.867188) net_send: could not write to fd=5, errno = 32 ------------------ Now as your revelation,I will try Diag.ParallelOverK true! Wish it good! REGARDS! ------------------ BEST REGARDS! Guangping Zhang ----------------------------------- Atom and Melecular Physics Physics and Electronics College Shandong Normal University Shandong,Jinan,China ------------------------------------ ------------------ Original ------------------ From: "Yusia Rosee"<[email protected]>; Date: Thu, Mar 11, 2010 10:17 PM To: "Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta";<[email protected]>; Subject: [SIESTA-L] About parallel mode in Transiesta Hello, Dear transiesta users and developers, I am very grateful that we can have access to Transiesta now. It provides us a very good tool for transport calculations. However, when I am reading the manual and doing some tests, I find that the parallelization can be done only over K points and the number of processors can not be larger than the number of K points. It means that for a very big system with nanowire as the electrodes where I want to use only 1 k point in the x and y direction, I can only do a serial calculation. It will be very time consuming and definely a bad news. In a usual Siesta calculation, the parallelization can be done either over K points or over orbitals. Even if we can not parallelize over orbitals, I think we should still be able to realize parallel over the energy integration. Thus we can still use many processors to speed up the calculations. How can we realize this (parallel energy integration), please? Sincerely, Yusia
