[wrfems] Old DX servers

Daniel Leins Daniel.Leins at noaa.gov
Wed Dec 8 07:27:09 MST 2010


Doug,

I haven't worked with the DX's yet, but there's a chance we might be. 
I'm curious as to what you find out.

That being said, I was under the impression that the DX's were 32bit 
machines, not 64. I could be totally wrong about that though.  I also 
thought they had 4 processors each. How many DX's are in your cluster?  
Have you tried running the benchmark case with ONLY the 4 DX's? I'd be 
curious to see what your results are without mixing different types of 
machines.

The DX's should all have gigabit NIC's, what kind of network issues are 
you dealing with?

Dan
WFO PSR

On 12/8/2010 7:16 AM, Douglas Tilly wrote:
> To all of the WFOs out there...
>
> Has anyone had success using the old DX servers in their clusters?  I 
> installed a fresh OS and have everything up and running.  I'm using 
> the core2 binaries due to another machine in the cluster.  When using 
> two machines with 18 processors, I get compute time for each timestep 
> of around 0.9 seconds.  When I use the old dx server with 4 extra 
> processors, it increases to around 2.3 seconds.  When I run sysinfo 
> (for whatever reason, I don't have cpuinfo anymore), it says to use 
> the p7 binaries.  So, when I use those, it's still slower than before 
> I replaced the machine that only had two processors.  I'm not sure how 
> this makes any sense.
> When looking at /proc/cpuinfo, I get:
>
> vendor_id       : GenuineIntel
> cpu family      : 15
> model           : 4
>
> which is the same as the old LX boxes that I'm trying to replace.  I 
> was hoping that the 64-bit servers would help speed it up.  Maybe I'm 
> just not using the correct binaries?  I'm also still investigating 
> some network problems along with a potential hardware issue.  Just 
> wondering if anyone else has tried using these dx servers and what 
> your experiences are.
> Thanks.
> Doug
>
>
> _______________________________________________
> wrfems mailing list
> wrfems at comet.ucar.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.comet.ucar.edu/pipermail/wrfems/attachments/20101208/c423909a/attachment.html>


More information about the wrfems mailing list