COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Dear Craig:
C00007 ENDMK
C⊗;
Dear Craig:

	By all means, determine the cost of using an IBM 370 to put a 3850
on the network, but my guess is that it will be far more cost-effective
to put it on one of the existing PDP-10s, especially one that uses the 3330
already such as ours or CCAs.  Ralph Gorin estimates that it could be
operating on our system two months after delivery of the hardware in a mode
where files on the 3330 would be in our file system and could be put in and
out of the system by FTP.  Thus TENEX users would need no new software.
It is that easy, because the 3850 is designed to act as a virtual 3330.
Besides the 3850 itself, our 3830 controller would have to be converted from
a model 2 to a model 3.  Gorin thinks we would also need another 3333 which
is the disk module with some additional control capability, we would probably
need to replace our present homemade channel by an SA-10 made by Systems
Concepts and might need one or two more 3330 modules to act as buffers, but
more likely we could get by with the four we have now since we could put
some of our own files on the 3850.

	While we would like to do this, it is more important that the network
have a 3850 than that we operate it.  An argument in favor of our operating it
or the M.I.T. AI Lab operating it is uptime.  As you know, hacker maintained
machines have much less downtime than professionally maintained machines,
because the "fix it when it breaks" method leads to less downtime than
scheduling substantial amounts of time for maintenance.

	The cost of the SA-10 mentioned above is $37K and the maximal additional
rental for IBM hardware mentioned above is about $3K per month, and so these
costs are minor compared to the cost of the $3850 itself.  I forget the rental,
but the price of the 35 billion byte minimal version is about a million.
Unless Marrill has tied himself into a knot with his 3330, he ought to be
able to provide the same service almost as cheaply.  I should say that I see
no additional personnel cost if we did it.

	As to how much we would spend on the use of a network 3330, I can't
rightly say.  Of course, the price for storage of information on it would
have to be lower than the price for adding more modules to our 3330 system,
even non-IBM modules.  Maybe we could go back to model 1 3330s if there were
such a 3850 and reduce the number of drives.  We now have 4 drives and they
come in pairs.  Maybe we could even get rid of our user disk pack and have
users keep their data on the 3850 and go to only two 3330 drives.  As an
extreme, we might even be able to go to an imitation 2314 instead of our
3330.  This also depends on how fast data can be transferred over the network.
Maybe you need to do a market survey.

	On the other hand, if there were a 3850 on the network at a really
low cost, say at not more than twice the rental of the bare 3850 to IBM,
we would be tempted to keep some really big files.

	I don't want to go into our KL-10 plans and our arguments for
time-sharing research now.