GOESnet rains satellite data -

The National Oceanic and Atmospheric Administration's Science Center in Camp Springs,
Md., had to pull the plug on its two IBM Corp. 4381 mainframes. They were choking on
satellite imagery flowing down from NOAA's newest geostationary satellites.

In their place, NOAA's Satellite Services Division has put in GOESnet, a network of
servers named for the pair of geostationary operational environmental satellites that feed
the network 48G of data every 24 hours.

Even NOAA can't get away from the food metaphors, calling its pair of frontline
dual-processor Silicon Graphics Inc. Challenge DMs the ingest systems, because they take
in raw data from the geostationary satellites.

The IBM 4381 mainframes were good input-output systems for NOAA's needs in the early
1980s, said Benjamin Watkins, chief of the Satellite Services Division.

But between them they had only 60G of disk capacity, and their processing speed of 16
million instructions per second couldn't keep up with a fourfold increase in data about
three years ago.

Only IBM 3090-class mainframes could have coped with NOAA's processing needs at that
point, Watkins said. NOAA's Interactive Processing Branch officials decided they needed a
new architecture to do the job.

"Our motto has always been: We serve those who serve others," branch chief
Michael Matson said. "We think of GOESnet as a server of servers."

GOESnet is a network of primary and secondary servers built to push satellite data out
to users within 30 minutes of the initial satellite scan. That's happening now with the
redundant SGI Challenge DM ingest systems attached to the primary servers via a private
Fiber Distributed Data Interface ring.

"If we have a catastrophic failure here ingesting the data, you're probably not
going to see satellite imagery on your TV at night," Matson said.

As for the mainframes that GOESnet replaced, only their I/O channels will be missed,
said one NOAA support staff member.

The primary GOESnet servers, a pair of IBM RS/6000 Scalable Processor three-node
systems, process and remap satellite imagery and sounding data.

Their 77- and 66-MHz IBM Power2 processors have math-handling capabilities superior to
those of current IBM PowerPC processors, said Brian Callicott, lead systems administrator
for GOESnet contractor PRC Inc.

The secondary servers, attached to the primary servers via a second FDDI ring, are
mostly IBM RS/6000 591 models. They produce and distribute NOAA's data products, the most
popular of which are wind speeds and directions, cloud tops, flash-flood forecasts, and
temperature and humidity data.

High-density winds, a new data product this fall, will be like every NOAA product,
requiring "more number-crunching, more memory and more storage," said NOAA
systems application team leader Susan Ladenheim.

To deliver these products to the primary user, the National Weather Service, and other
government and commercial users, NOAA's Interactive Processing Branch relies entirely on
noncommercial applications and source code developed and maintained by the University of
Wisconsin.

Commercial vendors don't provide the source code with their application software,
Callicott said.

The Interactive Data Analysis System, developed by university researchers, helps
weather analysts tell when a cloud formation is going to become a thunderstorm or whether
a tropical storm will turn into a hurricane.

Fortunately, that software could be recompiled on a variety of mainframe and
workstation processors, Callicott said, so NOAA was able to move from the mainframe to a
client-server network.

The move is costing NOAA's Satellite Services Division $8.2 million over four years,
ending in fiscal 1998, Matson said.

Part of the move has involved giving NOAA's geostationary satellite data users an
alternative to proprietary hardware known as the WideWord and Tower workstations.

NOAA has been replacing the proprietary workstations with standard Intel Pentium PCs
running IBM OS/2 Warp, and with Hewlett-Packard Co. HP 755, IBM RS/6000 590 and 591, and
Silicon Graphics Indigo workstations running Unix.

The agency is now ahead of its storage requirements, having two RS/6000 SP servers with
their own 432G storage banks. "If you want to get warm, stand behind them,"
Ladenheim said.

Each day, GOESnet now receives 60G of geostationary satellite data--the combined intake
from NOAA's GOES-8 and GOES-9, the European Meteosat-6 and Japanese GMS satellites. Before
long, GOESnet will also take in information from the Indian Insat and Chinese FY-2
geostationary satellites.

"Right now, we have a data gap over the Indian Ocean because we don't have the
Insat data in here yet," Matson said.

NOAA isn't finished with GOESnet. Over the next few months, it will add four 64-bit
Silicon Graphics Origin2000 symmetric multiprocessing servers to handle large data sets
from the new satellites coming on line.

"Initially, they'll run the programs single-threaded just as they are now,"
Ladenheim said. But NOAA is preparing to run its first multithreaded programs "and
get data out even faster."