[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

20020219: decoding multiple copies of Unidata-Wisconsin images



>From: Anthony James Wimmers <address@hidden>
>Organization: UVa
>Keywords: 200202192053.g1JKrYx27851 McIDAS-X AXFORM

Tony,

re: XCD decoding of grid files

>All the changes you made are exactly right. Now, about your
>questions:

re: did you know that MRF was going away

>I didn't know that, so thanks for the heads-up. If the MRF is leaving
>anytime before this summer, then I suppose that we can cancel it anytime
>on windfall, after I check to make sure we're not using it anywhere.

OK.

re: The conduit feed has grid 3 fields that you will most likely
be interested in (global 1 x 1 degree).  This is one of the grids
that gribdec.k was unable to decode until my recent work.

>Right. Grid 3 would be all we want.

OK.  We can attend to that after the McIDAS addendum AND after we
face the issue of upgrading your McIDAS distribution.

re: reduced XCD grid decoding so /p4/data has some breathing room

>It's been stable for several days now, so I think you're right. Now,
>there's what looks like a related issue to deal with. I thought that
>it would fix itself after we fixed the diskspace problem, but apparently
>it didn't. The problem is that the GOES images are coming in doubles:
>
>IMGLIST RTIMAGES/GE-IR.ALL
>Image file directory listing for:RTIMAGES/GE-IR
> Pos Satellite/         Date       Time      Center   Band(s)
>     sensor                                 Lat  Lon
> --- -------------  ------------  --------  ---- ---- ------------
>   1  G-8 IMG       19 FEB 02050  11:15:00    23   71 4
>   2  G-8 IMG       19 FEB 02050  12:15:00    23   71 4
>   3  G-8 IMG       19 FEB 02050  12:15:00    23   71 4
>   4  G-8 IMG       19 FEB 02050  13:15:00    23   71 4
>   5  G-8 IMG       19 FEB 02050  13:15:00    23   71 4
>   6  G-8 IMG       19 FEB 02050  14:15:00    23   71 4
>   7  G-8 IMG       19 FEB 02050  14:15:00    23   71 4
>   8  G-8 IMG       19 FEB 02050  15:15:00    23   71 4
>   9  G-8 IMG       19 FEB 02050  15:15:00    23   71 4
>  10  G-8 IMG       19 FEB 02050  11:15:00    23   71 4
>IMGLIST: done
>
>The problem manifests itself on our Weather Page by getting all the
>times wrong in our 24-hour image loops, so it's kind of important to
>fix.

OK.

>I noticed this problem the same time I noticed the diskspace problem, so
>I assume they're related.

Yes, they were.

>Do you think that I should try cleaning the
>spool? That's the only thing I can think of besides looking upstream for
>the glitch.

The problem turned out to be two invocations of LDM's pqact running
when only one should be.  The effect of this was that each pqact would
decode each image as it came in, so you would get two copies.  Two
invocations of pqact must have been caused by the disk filling up.
Also, there were some routines that had been running for several days
and really eating up CPU.

The fix for all of this was:

<as 'ldm'>
ldmadmin stop

<kill all 'ldm' processes that would not die>
ldmadmin start

After doing the above, the load average on windfall dropped from 7 to
less than 1.  Also, now only one invocation of pqact is running, so you
will only get one decoded image for each one ingested into the LDM
queue.

I did not remake the LDM queue as that did not seem to be necessary.
If you see other problems, we may want to revisit this.  In that case,
I would recommend editing ~ldm/bin/ldmadmin, and increasing the queue
size to 400 MB.

>Thanks,

Let me know if you see anything amiss tomorrow.

Tom