[wrfems] Clustering issues

peter.blottman Peter.Blottman at noaa.gov
Fri Sep 24 08:47:30 MDT 2010


OK, so that's a change back to the old way. No bigee especially since 
the days of many compute nodes that required the monitoring of many NFS 
mounts are gone.  I assumed you got around that problem in V3 because 
you were just sending node logging to /dev/null. Thanks for the heads up 
with V3.2.


On 9/24/2010 10:16 AM, Robert Rozumalski wrote:
> Kurt,
> Actually, I still recommend it and it will be required for V3.2 simply 
> because it's more trouble for me is I
> allow users to not do this.  One problem that will be encountered is 
> that since MPICH does I/O in the form
> of the rsl* files on each node, where will be no place to write the 
> files and MPICH will default to writing
> to the screen. That gets messy.
> Bob
> Kurt Mayer wrote:
>> You don't need an NFS mount? That'll save some grief the next time 
>> around.
>> On Thu, Sep 23, 2010 at 7:07 AM, peter.blottman 
>> <Peter.Blottman at noaa.gov <mailto:Peter.Blottman at noaa.gov>> wrote:
>>      Hi Mark,
>>     With the later versions of EMS (V3) you will not need to NFS mount
>>     directories between machines. You just need to ensure you are able
>>     to ssh between all your nodes without asking for a password.
>>     Passwordless ssh can be set up by exchanging the public keys
>>     between the nodes and appending them to authorized_keys within the
>>     users .ssh directory. A reasonable description of the process is
>>     described here: http://www.securitydocs.com/pdf/3385.PDF The EMS
>>     distribution will handle all the requirements for MPICH etc., so
>>     nothing needs to be installed there. The host file for each
>>     machine will just need to contain those machines, and their subnet
>>     IPs, listed that you are setting up in your cluster. I'm sure
>>     others in the group will fill in the details if required beyond my
>>     general description.
>>     Pete
>>     On 9/23/2010 2:10 AM, Mark Jarvis wrote:
>>          Hello:
>>         Is there a step by step guide out there about clustering and
>>         what MPICH2 actually needs?  For instance, what needs to be in
>>         the hosts file?  Do all the nodes need an NFS mount to /usr1
>>         on node1?  I was having problems previous with the NFS
>>         mounting, but figured out the problem there.
>>         I'm just trying to cluster with two workstations and the hosts
>>         file thing is really messing with netcheck.  Am not sure what
>>         needs to go in the hosts file.
>>         Mark @ Louisville, KY
>>         _______________________________________________
>>         wrfems mailing list
>>         wrfems at comet.ucar.edu <mailto:wrfems at comet.ucar.edu>
>>     _______________________________________________
>>     wrfems mailing list
>>     wrfems at comet.ucar.edu <mailto:wrfems at comet.ucar.edu>
>> ------------------------------------------------------------------------
>> _______________________________________________
>> wrfems mailing list
>> wrfems at comet.ucar.edu

-------------- next part --------------
A non-text attachment was scrubbed...
Name: peter_blottman.vcf
Type: text/x-vcard
Size: 214 bytes
Desc: not available
URL: <https://mailman.comet.ucar.edu/pipermail/wrfems/attachments/20100924/7c6e50db/attachment.vcf>

More information about the wrfems mailing list