To: Andy Edman <Andy.Edman@noaa.gov>,
 	bernard meisner <bernard.meisner@noaa.gov>,
 	kenneth johnson <kenneth.johnson@noaa.gov>,
 	Kirby Cook <Kirby.Cook@noaa.gov>,
 	preston leftwich <preston.leftwich@noaa.gov>
CC: "Stephen J. Lord" <Stephen.Lord@noaa.gov>
Subject: T254 implementation

Folks,

    Bernie alerted me that the poll currently taken by NCO should have
more background material to inform the users. I have decided to do that
from EMC's point of view.

    At the present time, the AVN run takes 60 minutes to finish 7.5 days
of forecast. This comes out to about 8 minutes per day of forecast. This
time was established on the phase II of the current IBM-SP computers.
Historically, the time per day of forecast for the global model has been
shrinking from 24 minutes per day 15 years ago to eventually the 8
minutes per day we currently maintain. We knew that we will not be able
to keep this rate if we implement the T254 on this machine. In our own
estimate done before CAFTI, we would be near the 60 minute mark if we
really load up the machine in the T170 portion of the forecast (from 3.5
to 7.5 day range). The NCO estimate is based on the addition of product
generation due to the larger file size which we didnot include and the
use of fewer nodes for the T170 portion of the run to preserve their
ability to play catchup when there is a problem with the system. We at
EMC is worried that the window of 8 minutes per day is not going to
leave us with anyroom to make further improvement in the global forecast
system in the future. As a comparison, the eta model has always enjoyed
the 24 minute per day window. The AVN has extended it's forecast length
and shrunk the window. We don't see us thriving in this environment.

    In order to increase resolution (it is important to do that for the
data assimilation and the short-range portion of the forecast), we have
to perform extensive tests. This is never possible on an older machine
when the resolution increase is aimed at the new machine. This means
that we will always go to a new machine with lower resolution than we
planned. The global model has not been aggressive with the resolution
increases over the years and we were able to reduce the runtime.
However, we do need to increase the resolution from time to time in
order to keep up with the other centers of the world (not to mention the
ECMWF which is now at T511). In the T254 change, we performed a very
long period of test and proved to ourself that there is significant
improvement in the performances of the suite. We would hope that the
change can be implemented now so that we can continue with our other
planned tests.

   Given that the new computer we plan to purchase is about 2.5 times
faster than current, we will make the window on the new machine if we
decide to wait. However, this will mean that we will have to do without
other changes we planned on the new machine. We may need more time to
ingest and analyze more and more new satellite data. We know that we
planned to make hourly output for the first 36 hours to provide eta like
meteogram files. We were also going to make .5 degree GRIB files for AWC
for global flight planning (also for the first 36 hours). We also have
planned to add more tracers (possibly aerosols or the eta cloud scheme).
If we stay at the same window, we will have to wait until the phase two
of the next machine to do most of these.

    The bottom line is we need a reasonable window to be allowed to make
our model runs and we feel that 8 minute per day is simply not
sufficient. We understand the need of forecasters to have timely
products and we have been doing our best to not let the window extend.
However, I think we have reached a point where we are going to be harmed
by the window. One of our strategy to avoid this was to make the 4-times
per day run be extended to make the products available on a continuous
bases. We were hoping that forecasters will take the latest available
model predictions and make updates based on them. There needs to be a
balance of the timelyness of product to the time needed to make a good
forecast. We feel that the current window will begin to hurt EMC's
ability to make improvement and hence our request for an extension of
the AVN window.

     Please feel free to provide feedback from the field.

Hua-Lu