McIDAS-OS2 IDD Leaf Node Implementation Plan
Tom Yoksas
June 1994
Introduction
In spite of continual encouragement by the UPC for movement to Unix, a substantial
number of sites (~30) continue to participate in Unidata with McIDAS-OS/2 platforms.
The reasons that sites may wish to continue to use OS/2 platforms generally fall
into two categories: a) platform cost; and b) ease of configuration and system
management. Unidata, through direction of its User and Policy Committees, has
accepted the need to continue to support these sites for the near to indeterminant
future.
This paper outlines strategies considered by Unidata for continuing to support
these typically-small sites as Unidata moves to the Internet Data Distribution
(IDD) method of disseminating weather data.
Support Options
The options recognized by Unidata for continued support of OS/2 McIDAS systems
are:
- movement of satellite broadcast of the Unidata-Wisconsin data channel to
the new Alden service by December 1, 1994
- use of 'mctingest' as an ldm "decoder" at IDD redistribution sites to feed
McIDAS-OS2 systems an Internet-based Unidata-Wisconsin channel
- development of an LDM OS/2 client that would receive data from an IDD redistribution
site
- a full port of the LDM to OS/2 coupled with the development of the LDM
client module mentioned above
- development and deployment of "IDD in-a-box"
The strengths and weaknesses of each of these approaches are presented.
Continued Satellite Broadcast of the Unidata-Wisconsin data channel
Unidata has considered moving the broadcast of the current Unidata-Wisconsin channel
from C band to KU band after the December 1, 1994 sunset of C band transmissions
by Alden. This move would cost $5300/mo as compared to the current $3750 being
paid for C band transmissions. Since the cost of this service is considerably
higher than it is currently, and since Unidata's goal is to provide data to its
community through the IDD, we do not consider this a viable long term option although
it may be useful during the period when sites not yet on the Internet get connected.
As presented elsewhere in this meeting packet, the Unidata cost to continue C
band broadcast of any data stream would be on the order of $20K per month, a cost
that Unidata can not afford.
Use of 'mctingest' to feed OS/2 leaf nodes
'mctingest', an LDM McIDAS ingester developed by Steve Emmerson of the UPC, is
built upon the freely-available FSU's Pthreads package for SunOS 4.1.x, DECThreads
for OSF/1 and Ultrix, and the builtin threads package that comes with SunOS 5.x
(Solaris). This ingester, which has been in use at Unidata since September, 1993,
functions in a similar manner to the new "product queue" version of the ldm that
is under development: products are assembled from a feed and then sent to a list
of clients by independent threads. The independence of the feed threads insures
that machines that suffer from a "slow link" problem do not affect others. Unidata
has successfully used mctingest to feed up to 7 McIDAS-OS2 and McIDAS-X machines
simultaneously here at the UPC and around the US. The experience gained from mctingest
convinced us that producing a highly reliable, data distribution system was feasible.
Like other ldm ingesters/decoders written by Unidata, mctingest can operate from
a variety of inputs: files, devices, and standard input. The ability to use standard
input allows mctingest to be used as an LDM decoder that gets its input from an
LDM PIPE action. This means that we already have a reliable method of feeding
McIDAS-OS2 workstations (AND McIDAS-X workstations not using the LDM) that are
connected to the Internet). The drawback of using mctingest with the IDD is that
the configuration mechanism for mctingest is completely different than that for
the LDM. This would make it harder for a relay site manager to maintain and upgrade
IDD configurations at his/her site. It also, on SunOS 4.1.x machines, requires
that users maintain current distributions of a non-Unidata package, FSU's Pthreads.
Development of an LDM client for OS/2
Even though we currently are able to feed McIDAS-OS2 workstations with mctingest
(described above), we feel that effort should be devoted to developing an OS/2-based
LDM client that would allow IDD relay sites to send the Unidata-Wisconsin channel
products directly to McIDAS workstations. This effort, which is estimated to be
about two programmer weeks work to get the code functioning reliably, would allow
for a consistent configuration method for IDD relay sites. Documentation and release
engineering will take at least as long as the code development so the total effort
is on the order of a month's work. We believe that simple and consistent configuration
and management of relay nodes will be key to successful operation of the IDD.
The limiting aspect of this approach is the McIDAS-only orientation of the solution.
It is likely that most sites will eventually want to take advantage of the variety
of data offered on the full IDD.
Full port of the LDM to OS/2
A longterm option for expanded support of McIDAS-OS2 platforms is the port of
the entire LDM to OS/2. Review of the facilities provided by the two Unidata-recommended
OS/2 network packages (PC/TCP for OS/2 from FTP Software, Inc., and TCP/IP for
OS/2 from IBM) indicates that a port of the LDM to OS/2 would most likely be possible.
What is not known, however, is the extent of the effort that this port would take.
We also question the advisability of a move like this based on the slow acceptance
of OS/2 in the PC computing community as a whole. Given that we are already shorthanded
for the forthcoming IDD deployment effort, we do not believe that this option
is viable at this time.
IDD in-a-box
The last option that we have been considering is the creation/deployment of a
single-purpose, inexpensive, Unix-based machine that would be used for IDD reception
(and possible relay) of data at sites lacking other Unix machines. The idea is
to put together a machine (e.g. a 486-based PC class machine or a low end Unix
box) with a standard configuration (e.g. Solaris for 486 or for a Sun Classic)
that would only be used for IDD tasks. This machine would necessarily support
NFS so that other machines at a site could access locally ingested/decoded data.
This machine would NOT run the X Window system nor be used for other research
or educational activities. In this way, we would be assured that system overloading
would not affect IDD data movement activities. We also feel that by limiting the
facilities used we can minimize the Unix experise needed to run the machine for
IDD purposes only. We are confident that a machine that would serve this purpose
could be assembled for a cost of less than $3000.