[Pw_forum] serial algorithm in parallel phonon calculation

[Paolo Giannozzi]

On Wednesday 23 August 2006 18:32, Eduardo Ariel Menendez P wrote:

> Looking at the ph.x output, when running in parallel with four CPUs
> (-npool 1) I found the following intriguing report
> ***********************
>      Calculation of q =   0.2500  0.2500 -0.2500
>
>      looking for the optimal diagonalization algorithm ...
>
>      a serial algorithm will be used
> ***********************

this refers to the "conventional" diagonalization, used in the
Davidson iterative diagonalization to diagonalize over a subspace.
It is an additional level of parallelization that may or may not be
convenient to have. 

>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
> 11020 eduardo   16   0  107m  20m  50m R 63.9  1.0  32:49.09 ph.x
> 11019 eduardo   16   0  107m  20m  50m S 21.6  1.0  11:22.56 ph.x
> and the same in the other dual node. Is it related to the algorithm? 

no, it isn't. Are the two processing running on different CPU's?
it looks very lousy.

Paolo
-- 
Paolo Giannozzi             e-mail:  giannozz at nest.sns.it
Scuola Normale Superiore    Phone:   +39/050-509876, Fax:-563513 
Piazza dei Cavalieri 7      I-56126 Pisa, Italy