[news.admin] /usr/spool/news file system and 4.[23]BSD

jerry@oliveb.olivetti.com (Jerry Aguirre) (03/19/88)

If you are running 4.[23]BSD and mount /usr/spool/news as a separate
file system you may want to consider tuning the minimum free space
threshold.  I have mine set to 0% and use other mechanisms to keep a
10% reserve.

The "fast" file system has better performance if you keep enough free
space that new files can be added without scattering them all over the
disk.  The default is to reserve 10% of the space as only useable by
root.

While good performance is desirable one must decide if it is better to
suffer a performance degradation or to run out of space and have
received news articles create truncated files.  If I receive a
unexpectedly large amount of news I would rather have news use that last
10% of the file system.  I can clean up later to restore the 10%
reserve.  Some remaining files may be fragmented but everything will get
expired in a few weeks anyway.

My nightly expire script does repeated "df" commands and increasingly
more severe expires until I have 10% plus 4 Meg. free.  The 10% is to
keep performance good on that file system and the 4 Meg is for the next
days news.  The difference is that if I ever get 8 Meg of news I will
still be able to handle it.  Looked at another way I can tune my reserve
space requirement to the average rather than the peak value.

I would not recommend this unless /usr/spool/news is on its own file
system.
				Jerry Aguirre

mouse@mcgill-vision.UUCP (der Mouse) (04/06/88)

In article <18368@oliveb.olivetti.com>, jerry@oliveb.olivetti.com (Jerry Aguirre) writes:
> [suggests tuning the free space reserve on the news spool filesystem,
>  if your system has the capability]

> My nightly expire script does repeated "df" commands and increasingly
> more severe expires until I have 10% plus 4 Meg. free.

We do it a different way.  Our news comes in via nntp, active receive
(ie, we reach out and take it rather than their reaching out and giving
it to us).  We do hourly df commands and shut down the nntp program
which does news receiving when the usage figure hits 91%.  (It also
sends mail to usenet, so I can take whatever action seems appropriate,
and it also turns nntp back on when the usage figure goes back to 89%
or lower.)

(By the way, the nntpxfer program in the nntp 1.5 distribution is
incredibly inefficient.  Don't run it verbatim unless you like doing
one syscall per byte of news transferred.)

> I would not recommend this unless /usr/spool/news is on its own file
> system.

Gee, whyever not? :-)

					der Mouse

			uucp: mouse@mcgill-vision.uucp
			arpa: mouse@larry.mcrcim.mcgill.edu