Want Opinions Of Mainframe Versus UNIX For Client-Server Applic

[Will E Estes]

I would like some opinions and/or test results that would
give me a basis for comparing performance of a client server
application under one of two configurations.  In both
configurations the client side is a 386 PC running MS-DOS
and connecting to the host using a 2400 bps modem.

Configuration One:  Server is an IBM mainframe running
VM/CMS accessed through a protocol converter or,
alternately, accessed through a TCP/IP Ethernet and router
that supports asynchronous TCP/IP access.

Configuration Two:  Server is a multi-processor UNIX machine
(e.g., a Sequent S81 with 20 386 processors) accessed
through direct modem connections or, alternately, accessed
through a TCP/IP Ethernet and router that supports
asynchronous TCP/IP access.

The application is similar in concept to CompuServe's
Informatino Manager (CIM) or Connect Inc.'s Connect
software:  the user interface is on the PC and E-mail,
conferencing, and database engines are on the host side.
The protocol used to exchange messages between client-server
may be built from scratch (in the case of using protocol
converters or direct-connect modems) or based on something
standard like TCP (in the case of an Ethernet attachment to
the host).

I have been told that the application's performance is
likely to be poor under Configuration One if we use the
protocol converter, but no one has been able to quantify the
response time difference.  Does anyone have any opinions
regarding the four options?  (i.e., 1) IBM host w/protocol
converter, 2) IBM host w/Ethernet connection and asynch TCP
router, 3) UNIX host with direct modem connections, 4) UNIX
host w/Ethernet connection and asynch TCP router)

I'm also interested in anyone's thoughts on how asynch
TCP/IP (e.g., SLIP or PPP) is likely to perform over on a
2400 bps line in both configurations, and are there
preferred alternatives?

Thanks,
Will Estes             (apple!cup.portal.com!Will)