🚀 go-pugleaf

In article <1992Dec18.150228.8274@cis.uab.edu>, hyatt@cis.uab.edu (Robert Hyatt)
writes:

|>>Consider any application written for a line mode terminal using FORTRAN I/O.
|>The
|>>portability of these programs was pretty much setttled a long time ago. On
|>the
|>>other hand any code using a windows system soon becomes a porting hazard. On
|>>UNIX there are even difficulties in porting code from one UNIX system to
|>another
|>>of the model and from the same vendor. This is because sysops tend to want
|>to
|>>split up large libraries amongst disks and UNIX does not provide a mechanism
|>>like logical names which allow a user to virtualize a disk system to make it
|>>look like something else. Logical links don't count in this regard, they
|>affect
|>>the system environment.
|>>
|>
|>I don't understand the "logical links don't count"......  we use 'em for
|>exactly this purpose.  When Sun puts something somewhere that we don't
|>like (don't have room in that partition, etc.) we move that "thing" and
|>symbolic link to it.  We've never had any problems with X, openwin, etc.
|>in this environment.....

Try using them without system privs. 

There is no way of making a UNIX machine look as if it configured in one way to
one program but in quite a different way to another program. If you have more
than 20 people on a cluster you will realize why it is impossible to give root
privs to everybody. Even if people know what they are doing it is not acceptable
to have one person interfere with the environment of another person.


|>>
|>>UNIX is a single user system too. The only evidence of a multiuser capability
|>is
|>>
|>
|>This is news to us.  We have been using unix boxes for multiple users for 15
|>years.  Unix is as multi-user as any O/S I have used.  (this includes running
|>X over async modem lines if you want....)  We have students doing program
|>development, email, compiling, remote machine access, distributed processing
|>using domain sockets and/or RPC's, etc.

Having beds for rent does not mean that a bordello is a hotel.

A multiuser machine should be just that, it should allow multiple users to
access it without interfering with each other. We have already noted the
difficulties caused by the file system. In addition to this inadequate resource
monitoring means that the scheduler is practicaly useless, any user can simply
spawn tasks to increase the proportion of CPU alloted to their needs. 

If your definition of multiuser capability is low enough to incorporate UNIX
then W/NT will come close. From multiprocess to multiuser the transformation is
mainly in providing a more sophisticated filestore. Untill there are
applications to support the new file store multiuser W/NT will be irrelevant.
Once that happens you can be pretty sure that Microsoft move in that direction.
If they don't Borland, Norton or someone else will.


--

Phill Hallam-Baker