Path: utzoo!utgpu!water!watmath!clyde!cbosgd!ucbvax!ucbarpa.Berkeley.EDU!fair
From: fair@ucbarpa.Berkeley.EDU (Erik E. Fair)
Newsgroups: news.admin
Subject: Re: How long are you keeping news these days?
Message-ID: <22409@ucbvax.BERKELEY.EDU>
Date: 6 Jan 88 10:19:24 GMT
References: <3899@tut.cis.ohio-state.edu>
Sender: usenet@ucbvax.BERKELEY.EDU
Organization: USENET Protocol Police, Western Gateway Division
Lines: 57

Disk space has been a problem on ucbvax last fall for the first
time in years. We run netnews (both the spool and lib directory)
out of an Eagle's "h" partition, which is 140Mbytes.

policy prior to August of last year:

	keep all netnews for 30 days
	run expire once a week (early monday morning)

As of the first of August, we ran out of inodes.

I started running expire every night, because we couldn't afford
a week's worth of slop in used inodes. This was OK until September.
Then we ran out of inodes again, so I cut the expire time back to
25 days, and the daily script mails me a "df -i" every night after
expire completes, so that I can keep an eye on it.

The inode count has lately been floating at between 2500 and 4000
free, although the christmas - new year's week has brought that
back up to almost 10,000 free. I shudder to think what will happen
when everyone gets back from vacation.

Disk space has been less of a problem; I keep between 10 and 30 Mbytes
free. UUCP has its spool elsewhere, so UUCP feed queues aren't taking
away from that. NNTP (most of what we do these days for news transfer)
takes articles right out of the spool, so that's no problem either.

If I had my druthers, I'd expire the soc, rec, and talk newsgroups
in two weeks, rather than for 25 days and I'd let the rest of them
back out to 30 days expire, but right now that would mean running
expire twice every night, and we can't afford that either; ucbvax
is just too busy. I'd run the supposedly stupendous C news expire
(which, aside from blinding speed, allows for selective expire
times on a per-group basis), except that it produces a non-B news
compatible history file, and NNTP as currently written can't deal
with that. Someday.

In the mean time, I watch the daily "df -i" that comes in my morning
mail, with one hand poised on the expire-time dial, if the inode
or disk space free counts get too low...

Here's this morning's report:

Date: Wed, 6 Jan 88 02:08:07 PST
From: usenet@ucbvax.Berkeley.EDU (USENET News Administration)
Subject: USENET disk space report

Filesystem    kbytes    used   avail capacity iused   ifree  %iused  Mounted on
/dev/hp1h     140564   89593   36914    71%   27576   11336    71%   /usr/spool/news

	Erik E. Fair	ucbvax!fair	fair@ucbarpa.berkeley.edu

P.S.	Yes, I know, I could backup the filesystem and mkfs with more
	inodes, but that would eventually leave me with disk space
	problem instead of lack of inodes - I can basically take
	my pick on this one, AND ucbvax would have to come down to
	single user mode for a while we do that...