Re: AWIPS II and EDEX/CAVE info

Michael,

Thanks for all of this info, much appreciated!

Pete


On 02/15/2013 03:22 PM, Michael James wrote:
> Hi Pete and Kyle,
>
> > We are working up a proposal for the 2013 Unidata Equipment grants, 
> > and would like to base part of it around setting up a prototype EDEX 
> > server and CAVE clients to run in our computer classroom. I'm having 
> > some trouble finding information about machine specs for both an EDEX 
> > server and the CAVE client, and I'm wondering if you could provide 
> > some guidance.
> >
> > What kind of processor/memory/disk requirements should we be looking 
> > at for an EDEX server? Would something like a dual processor Opteron 
> > 4228HE (2.8 Ghz 6 cores each) with 32-64 Gb RAM and a few Tb of disk 
> > space be about right?  We might want to use this machine to also do 
> > some background image generation for maps for the web. How 
> > CPU/memory/disk intensive are the EDEX server processes on the machine?
>
>
> The specs of the Opteron look good, similar to what I'm using for 
> testing at the UPC.  I have only two machines to test EDEX, one is 16 
> cores (threaded dual-quad core) at 2.27 GHz with 24 GB RAM; the other 
> 8 cores with 6 GB RAM.
>
> The difference in performance between these two machines is not that 
> significant.  most of the lag in the system when a high volume data 
> feed is pushed through it comes from the EDEX decoders (as spring 
> plugins for the JVM) exceeding their in/out limit, causing a backup in 
> the message queue.
>
> So those specs are around what I'm working with, and I feel like such 
> a system is adequate to process nearly all of the relevant feeds and 
> product types.  The big question mark these days is can we tweak EDEX 
> enough to process 100% of the NEXRAD3 feed and the high-res models on 
> CONDUIT.  A fix for this is not known yet, but from my experience with 
> AWIPS II I get the feeling that the solution is not going to be 
> "bigger system" but "smarter system".
>
> > Our computer classroom currently has several Xeon W3505 2.53 Ghz 
> > machines with 6 Gb of RAM and NVIDIA Quadro FX 580 video cards w/512 
> > Mb RAM each, and some older machines running Pentium 4 3.2 Ghz 
> > processors, 4 Gb of RAM and NVIDIA Quadro FX540 video cards s/128 Mb 
> > of memory each. Would these be sufficient to run a CAVE client on each?
>
> I think the low end of the graphics card memory that has been tested 
> at NCEP offices was 256 MB, so the the FX 580 should be fine, but I 
> don't know how well the workstations with 128 MB would work.
>
>
> > We have a total of 15 workstations in our classroom - would a single 
> > EDEX server be sufficient to host that many CAVE thin clients, or 
> > should we plan for more? My perception from the descriptions that I 
> > have been able to find out on the web is that most of the processing 
> > power on the EDEX server goes toward churning through the data and 
> > generating products for the CAVE clients to use, and that the CAVE 
> > clients themselves are not a huge tax on the server. Is that right? 
> > If so, perhaps the difference between a single CAVE client and 
> > several CAVE clients hitting a single EDEX server isn't that large?
>
> You are right about EDEX doing most of the work, but the data sent 
> after a request from CAVE to EDEX is contoured and displayed by the 
> CAVE client using the GPU.  The thin client is a little different, as 
> it tells EDEX to construct the contoured image on the server and pass 
> that back to the client, sort of like a scene graph.
>
> > Thanks for any information that you can provide.
> >
> > Also, if/when the AWIPS II software can be made available to 
> > universities for testing, we would be very interested in 
> > participating in the test group.
> >
> > Thanks,
> > Pete
>
>
> I anticipate we will have a ready beta in the next couple of months 
> and we can add you to the pool then.  We ask all of our equipment 
> award winners who submit a proposal including an EDEX server to 
> participate.  I'm waiting for a new version from NCEP to test in the 
> coming weeks, and then more source code from the NWS-Raytheon baseline 
> which includes a new grib decoder.  I want to have that grib decoder 
> optimized by the time we release the beta.
>
> Let me know if you have any other questions about AWIPS II.  Your 
> hardware looks good, and should work well for AWIPS II if you have no 
> qualms about running RHEL or CentOS 5 instead of 6 (which is being 
> worked on).
>
> Michael


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden