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

20030704: upgrade to LDM-6 at NMT (cont.)



>From: "D. J. Raymond" <address@hidden>
>Organization: NMT
>Keywords: 200307041809.h64I9TLd018509 IDD LDM-6

Hi Dave,

>I just started rtstats for you.

Thanks!  I really appreciate you reporting stats because it helps
us keep track of IDD performance.

No, on to ingest performance on heron: it is very bad at the moment:

http://www.unidata.ucar.edu/cgi-bin/rtstats/iddstats_nc?HDS+heron.nmt.edu

The differential latency from heron to yin.engin.umich.edu can be seen
by doing the following:

- open

http://www.unidata.ucar.edu/staff/chiz/rtstats/siteindex.shtml?heron.nmt.edu

- click on the 'topology' link for HDS:

http://www.unidata.ucar.edu/cgi-bin/rtstats/iddstats_topo_nc?HDS+heron.nmt.edu

- click on the yin.engin.umich.edu link:

http://www.unidata.ucar.edu/cgi-bin/rtstats/topolatency?heron.nmt.edu+yin.engin.umich.edu+HDS


The time series plot from the last link shows that essentially all of
the latency you are seeing is in the link from heron to yin.

Since the latency plot for IDS|DDPLUS is exactly the same for HDS, I
imagine that your ldmd.conf file has a single request line that looks
more or less like:

request WMO     ".*"    yin.engin.umich.edu

The latencies being as high as they are (ranging up to 3600 seconds)
warrants you splitting your feed request into at least two:

request IDS|DDPLUS      ".*"    yin.engin.umich.edu
request HDS     ".*"    yin.engin.umich.edu

If your setup is like a number of others, this split will at least drop
the latencies for the IDS|DDPLUS observational data down to low values.

Another possibility is that the clock on heron is off by as much as
1000 seconds.  We have been advising all sites running LDMs on PC OSes
(e.g., Linux, Solaris x86, and FreeBSD) to set their clocks at least
once per hour.  One way that folks are doing this is by running
'ntpdate' out of 'root's cron.  The University of Georgia uses 'rdate'
in much the same way.

After you split your feed, we will be able to see if your clock is, in
fact, off since our expectation is that the latencies for the IDS|DDPLUS
feed should drop to near zero values.

If you continue to have high latencies after splitting your feed requests
and adjusting your clock if needed, we will want to move you to a closer
IDD relay node.

I will keep an eye on your latencies for the next few days, but please
let me know when you have finished splitting your feed request.

Thanks in advance...

Tom