gnodes & login limits

[Kent Black,L08,640,7754072]

In article <14200 at reed.UUCP> kab at reed.UUCP (Kent Black) writes:
>
>We* have 3100's, just upgraded to Ultrix v3.1/UWS 2.2.
>	First build was
>		maxusers	5	# conserving a little space
>		physmem		12
>		no QUOTA

Thanks for the helpful mail, problem is understood: Do not compile
without QUOTA!

mellon at decwrl.dec.com: "... without the QUOTA option, the quota() call
to determine how many users can log in will return -1, since the quota
system call isn't in the kernel anymore."  Similar messages from
rusty at garnet.berkeley.edu & farhad at tehran.stanford.edu so far, I
assume more will come; forgive me if I do not thank you individually.

I thought I had another kernel running without QUOTA and had ruled
this out, but it turns out I had my own login(1) running there, which
I had forgotten.

Will DEC at least document this behaviour somewhere obvious (like
conf/GENERIC) unless or until it gets fixed?

Farhad suggested compiling with maxusers = 16.  These systems already
spend too much of there lives swapping processes, though, with only 12M
of memory.  For amusement I watch half a dozen processes swap (and
seconds pass) before a new xterm will start.  Anybody else trying to
run 3100's with this little memory?  (Vishnu has 24M and has a far
happier life.)

I am still interested in any info on ideal gnode limits -- should
there always be some free, or do they get recycled; i.e., is NGNODES a
hard limit as I think it is.  We have seen one crash with "out of
gnodes" and "file table full" panics; does this always cause a crash,
or did I just get lucky?  (We have several times noticed a paucity of
gnodes, less than 5 available, so I assume that the limit has been hit
a few times, but only one crash; I have since raised the count
significantly.)

Thank you all again,
-- kab