Re(2): [GE users] Comparison of MPI Job Runtimes with and without SGE

Reuti reuti at staff.uni-marburg.de
Thu Aug 12 00:22:00 BST 2004


    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

Waseem,

>     I am using ch_p4. mpich-1.2.5.2,.. rsh. I did not know about the 
>usefulness of serv_p4.i will defintely try it out.
>I would assume that both SGE and MPICH spawns jobs in similar fashion 
>through rsh. In fact tight integration may pay a little more cost 
>because of the extra cost of exceuting the rsh wrapper. is'nt it?

the wrapper is only a small script starting qrsh. Then qrsh will initiate the 
creation of the rshd via execd (this may be the fast step), and contact the 
rshd via rsh but with an already setup port. Furthermore, the command is not 
transmitted with the rsh, instead also tranferred to the spool directory of the 
nodes, and the qrsh_starter is just looking in the directory which name is send 
with this command.

The normal rsh will have to issue the command, the slave will see the request 
and give it to inetd, the inetd will act according to the red in inetd.conf 
what to do with this request. Oh, it's an rsh, I have to start an in.rshd 
session (this will look into pamd for the valid logins I guess), finally we got 
it. But that this difference is so big...Can you remove the option -L from 
inetd.conf in the call to in.rshd (logging of the rsh attempt)?

Reuti

BTW: Is your 100 sec job just making one rsh call if run in parallel, or many?

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list