[CPMD-list] TCPU

[Axel Kohlmeyer]

>>> "HW" == HW Sheng <hwsheng at jhu.edu> writes:

HW> This is a multi-part message in MIME format.

HW> ------=_NextPart_000_0005_01C268D8.60999DE0
HW> Content-Type: text/plain;
HW> 	charset="iso-8859-1"
HW> Content-Transfer-Encoding: quoted-printable

HW> Greetings, all.

HW> I am runing an ab intio job on a Linux/Alpha cluster. (10 dual Ev 6 =
HW> alpha processors + 3com network switch + LAM). I noticed that the TCPU =
HW> of the output is around 20 seconds (per step), but the real time almost =
HW> doubles the number. Does that mean that half of the running time was =
HW> wasted on the data transfer? Did this happen to you before? I used =
HW> netPIPE to check the network performance, and nothing unusal was =
HW> detected. Thanks for your input.

HW> Howard Sheng


hi howard!

this is to be expected. running cpmd in parallel needs an extreme amount
of communication bandwith. that means on a (i assume) 100MBit network
you will have to wait rather long for the communications.  

TCPU gives you the amount of cputime used per time step. since the lam
mpi library does not consume cpu time while waiting for communications
(which is a big advantage over the competing mpich library, IMO), you
are _only_ seeing how much time was used for the computations in cpmd.

from my experience - we are running cpmd in parallel over 100MBit
ethernet (rarely), Gigabit ethernet (routinely) and SCI (predomninantly
now) - you can get a reasonable performance over ethernet only for not
too large problems and a small number of compute nodes (up to 4 dual cpu
machines, up to 8 single cpu machines for gigabit ethernet, 2/4 for
100MBit ethernet). if you want to run cpmd on a larger number of
machines efficiently, you need to have a high-speed (and high-price)
network like SCI or myrinet.

all the best,
    axel kohlmeyer.

HW> ------=_NextPart_000_0005_01C268D8.60999DE0--

HW> _______________________________________________
HW> CPMD-list mailing list
HW> CPMD-list at cpmd.org
HW> http://www.cpmd.org/mailman/listinfo/cpmd-list



--

=======================================================================
Axel Kohlmeyer       e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum                   http://www.theochem.ruhr-uni-bochum.de
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.