Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!think.com!mintaka!spdcc!rbraun
From: rbraun@spdcc.COM (Rich Braun)
Newsgroups: comp.unix.sysv386
Subject: Re: Number of users on SCO
Message-ID: <8128@spdcc.SPDCC.COM>
Date: 28 Jun 91 17:37:06 GMT
References: <665@fudd.dataco.UUCP> <1991Jun28.084619.18114@kithrup.COM>
Organization: Kronos Inc., Waltham, Mass.
Lines: 29

campbell@bung.UUCP (Duncan Campbell, VOR) writes:
>>Howdy, can anyone tell me how many user accounts SCO SystemV 3.2.2
>>can support?  100? 1000? more? less?  Just accounts, not simultaneous...

sef@kithrup.COM (Sean Eric Fagan) writes:
>Assuming infinite disk space, it should be somewhere around 
>	a)  65435:  2**16 (65536) - 1 - 100 (you shouldn't use a UID
>		of less than 100; they're "reserved" by AT&T)

A number of years ago, I maintained a DEC-10 system at the University of
Delaware with over 9,000 user accounts.  The limitation isn't the size
of the integers used to identify users, but rather the speed with which
the password and/or other accounting files can be scanned.  The DEC-10
O/S was shipped with source code, so I modified the LOGIN program to
do a binary-tree search of the password file.  Most Unix systems, including
SCO I'm sure, do a linear search of /etc/passwd (in addition, SCO keeps
user records in the /tcb/files/auth tree) before asking for the user's
password.  This can become painfully slow if the file gets large and the
system isn't fast.

Therefore you need to correspond with someone who is now using SCO Unix
to manage a large number of users, rather than to simply be told that
the physical maximum is 64K or thereabouts.  A 100-entry password file,
by the way, would definitely not be considered "large", and I suspect
a 1000-entry file wouldn't make things too slow--but I don't speak from
personal experience with large password files on a 386.  The 'grep'
program has no problem with multi-thousand-line files, though.

-rich