Re: 19990302: LDM hardware configuration

Bob,

Here's an answer about replacing an IRIX box with a pc but the ideas about
the LDM architecture are correct for any box, ie Octane.

Robb...


From address@hidden Tue Mar  2 14:57:15 1999
Date: Tue, 2 Mar 1999 12:34:10 -0700
From: Mike Schmidt <address@hidden>
To: Peter Neilley <address@hidden>, address@hidden
Subject: Re: Replacement IITA machine

Peter,

Here's my .02.  I recall that IITA suffered as much from a sluggish
I/O system as it did from lack of memory.  I'd prioritize as follows
(based on experience with IITA and our LDM systems);

 - fastest I/O subsystem possible (ultra2/wide/lvd/10Krpm scsi disks)
     with enough disk capacity to meet your needs - depending on your
     decoding needs, you might consider low end RAID
 - as much memory as possible - 768 MB (or enough for LDM queue)
 - dual 400+ MHz processors (any Pentium II should work fine)
     Buying the latest Intel processors (Pentium III/Xeon) is too
     expensive to justify within the budget constraints you mention
 - fast network interface (100baseT is cheap anyway)

It's easy and relatively cheap to add PC commodity disk and memory later
if you underestimate -- presuming all the slots/drive bays are not filled.
Changing the I/O bus (and the disks) or the CPU's are much more costly.

For reference, our current LDM system is a Sun Ultra 2 with 2x300MHz CPU's,
1GB memory, ultra wide scsi with ~11 ms access disks (no RAID) and 100baseT.
Both the disks and the memory are always hammered.  The CPU's run along about
50-70% idle, and the network load is very low.

mike


> 
> ------- Forwarded Message
> 
> >To: address@hidden
> >cc: address@hidden
> >From: "Bob Creasey" <address@hidden>
> >Subject: ldm ?
> >Organization: .
> >Keywords: 199903022129.OAA24383
> 
> Hello,
> 
> I am considering a dual processor SGI Octane (IRIX6.5 512mb mem) to run our 
> LDM
> and product generation.  I was wondering if you had an opinion on whether a
> dual processor system would work well with the ldm environment and products or
> if the IO required made a single processor just as effective?  Normally, I
> would just keep with a single processor and fan out to other machines when
> loading became an issue at peak times.   Our current system is a single
> processor ELAN and it never sees a dull moment...I have to shoe horn in any 
> new
> tasks during processing gaps to keep it from blowing up.
> 
> Anyway I would be interested in your comments?
> 
> Thanks,
> Bob
> 
> -- 
>  Bob Creasey                         (  +*))   (+ * )_ 
>  Naval Postgraduate School          ((* - * )(++,, , )) 
>  Dept of Meteorology Rm 254 MR/CY (( ,- *-  ,, -,- ), ))__
>  589 Dyer Rd.                    (_______________________))
>  Monterey Ca 93943-5114              \   \     \\\\\\\    
>  address@hidden               / \  /\_    \\\\\\   
>  Work Phone: 1-831-656-3178           / / / \    \\\\\      
>  DSN: 878-3178 FAX: 1-831-656-3061      
> 
> 
> ------- End of Forwarded Message
> 

===============================================================================
Robb Kambic                                Unidata Program Center
Software Engineer III                      Univ. Corp for Atmospheric Research
address@hidden             WWW: http://www.unidata.ucar.edu/
===============================================================================