ralph@ralmar.UUCP (Ralph Barker) (04/15/89)
We have a number of sytems running System V, Release 3.0 which are
networked via Ethernet and TCP/IP. We are considering using RFS
as a means of "distributing" users and processor loads between
the machines, and I would appreciate comments, insights or horror
stories from others who have implemented similar designs.
The basic architecture of the "distributed" system would consist of
several classes of machines, each class being dedicated to a particular
generic function:
1. Execution Servers - machines holding identical copies
of application programs. Applications include
various "office automation" products, including
Uniplex, which finctions as the user's login
shell.
2. Home Servers - machines acting as the "home" for users
and their files, but NO applications
3. Special Purpose Servers - machines dedicated to "special"
functions which are more risky to run over the network
or are separated for security reasons
or other practical considerations:
- Database (Informix)
- UUCP, SNA etc (external communications gateways)
- Print servers
Our plan is to cross mount the appropriate file systems via RFS so that
users could continue to work, even if one or more of the Execution
Servers were to crash. Aside from "momentary" interuptions,
our objective is to create an environment in which the only time
a user would be "out of business" would be if their own Home Server
were to go down.
QUESTIONS: 1. Has anyone implemented a similar system architecture?
If so, what kind of problems were encountered in "real"
operation? (e.g. network congestion levels, RFS problems)
2. Has anyone THEORIZED about such an implementation?
If so, what thoughts or insights can you share?
--
Ralph Barker, RALMAR Business Systems, 640 So Winchester Blvd, San Jose,CA 95128
uucp: ...{pyramid, sun, uunet}!amdahl!unixprt!ralmar!ralph
or, attmail!ralmar!ralph Voice: (408) 248-8649