[SCore-users] 2 swiches, one server.

kameyama at pccluster.org kameyama at pccluster.org
Wed Jul 31 16:57:58 JST 2002


In article <200207310725.g6V7Ppf02906 at zeralda.streamline.com> Nick Birkett <nrcb at streamline-computing.com> wrotes:
> comp001.streamline     HOST_1 network=myrinet2k,ethernet,shmem0,shmem1 group=
> _scoreall_,MYRI,ETHER,SHMEM smp=2 MSGBSERV
> .......
> .......
> comp127.streamline     HOST_127 network=myrinet2k,ethernet,shmem0,shmem1 grou
> p=_scoreall_,MYRI,ETHER,SHMEM smp=2 MSGBSERV
> 
> /* 4 hosts - development cluster */
> comp128.streamline     HOST_128 network=myrinet2k-2,ethernet,shmem0,shmem1 gr
> oup=_scoreall_,MYRI,ETHER,SHMEM smp=2 MSGBSERV
> comp129.streamline     HOST_129 network=myrinet2k-2,ethernet,shmem0,shmem1 gr
> oup=_scoreall_,MYRI,ETHER,SHMEM smp=2 MSGBSERV
> comp130.streamline     HOST_130 network=myrinet2k-2,ethernet,shmem0,shmem1 gr
> oup=_scoreall_,MYRI,ETHER,SHMEM smp=2 MSGBSERV
> comp131.streamline     HOST_131 network=myrinet2k-2,ethernet,shmem0,shmem1 gr
> oup=_scoreall_,MYRI,ETHER,SHMEM smp=2 MSGBSERV
>
> pm-myrinet.conf is the standard conf for a 128 port switch and pm-myrinet.2co
> nf is a conf for 8 port switch.
> 
> Is this correct I wonder ? 

Yes, this is correct.
Please forget restart scoreboard and msgbserv.

I recommended to create different group name for comp128-comp131.

If you want to run scored on this 4 hosts,
you need to specify the hosts to run scout or scored.
If group name is not created, you must run:
    # scout -g comp128+comp129+comp130+compcomp131
    # scored
or
    # sc_watcfh -g comp128+comp129+comp130+compcomp131 ...

If you create group for comp128-comp131 (for example develop), you can run:
    # scout -g develop
    # scored
or
    # sc_watcfh -g develop ...

if you plan to add second cluster hosts, you may run separate msgbserv.
You can run multiple msgbserv one one hosts (with different port number),
then you want to add hosts in develop cluster, you must restart scoreboard
and msgbserv for development cluster.
You don't need to restart msgbserv for main cluster in this case.

                       from Kameyama Toyohisa



More information about the SCore-users mailing list