[Wrf-users] Help! computing experts
xingang.fan at wku.edu
Thu Mar 18 13:59:19 MDT 2010
Thank you and all who have replied. There were several colleagues who
have provided very helpful information and tips.
I am still collecting more detailed information on the configuration,
mainly focusing on getting a Dell HPC.
Xingang Fan, Ph.D.
Dept. Geography & Geology
Western Kentucky University
1906 College Heights Blvd, #31066
Bowling Green, KY 42101-1066
Email: xingang.fan at wku.edu
michael bane wrote:
>> Message: 2
>> Date: Wed, 17 Mar 2010 15:34:13 -0500
>> From: Xingang Fan <xingang.fan at wku.edu>
>> Subject: [Wrf-users] Help! computing experts
>> To: wrf-users at ucar.edu
>> Message-ID: <4BA13CC5.1000806 at wku.edu>
>> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>> Is there anybody out there who has experience of putting together a
>> small HPC cluster to run WRF model? I would appreciate it very much if
>> you could give any suggestions.
>> GOAL: I got an empty room and some funding (12-15K), and hope to have
>> WRF modeling results from there. So, what computing hardware and
>> software should I put into that room?
>> I have seen that DELL has different PowerEdges, but have no ideas
>> which one is the most appropriate one for our tasks. I hope to have
>> least 32 cpus, some disk space (~10TB). But to configure a small HPC
>> computer out of those DELL options seems too hard for me. Please share
>> your experience and help me out!
>> -- Xingang Fan, Ph.D.
>> Assistant Professor
>> Dept. Geography & Geology
>> Western Kentucky University
>> 1906 College Heights Blvd, #31066
>> Bowling Green, KY 42101-1066
>> Phone: (270)745-5980
>> Email: xingang.fan at wku.edu
> I've not noticed any other replies but I'd suggest you seriously
> consider investing in a rack and fast interconnects (InfiniBand,
> gigabit Ethernet), then populate the rack with various compute and
> disk nodes as you can afford (of course don't forget suitable air con
> too). This will give you some h/w upgradability and the interconnects
> will help with the parallel performance. It partly depends on what
> size model you want to run - if it's a small domain and fits on 24
> core then a single machine with 2 sockets of Magny Core chips would
> I'd seriously consider the disk side too if you're doing a few runs...
> or have somewhere to move the many GBytes of data off to
> Dr. Michael K Bane
> Senior Development Officer
> RESEARCH COMPUTING SERVICES
> Univ. of Manchester M13 9PL
> Wrf-users mailing list
> Wrf-users at ucar.edu
More information about the Wrf-users