Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/5/84; site tove.UUCP
Path: utzoo!linus!philabs!prls!amdimage!amdcad!decwrl!greipa!pesnta!hplabs!qantel!dual!lll-crg!gymble!umcp-cs!tove!israel
From: israel@tove.UUCP (Bruce Israel)
Newsgroups: net.misc
Subject: Re: Please help!
Message-ID: <305@tove.UUCP>
Date: Mon, 5-Aug-85 02:05:45 EDT
Article-I.D.: tove.305
Posted: Mon Aug  5 02:05:45 1985
Date-Received: Mon, 19-Aug-85 06:44:47 EDT
References: <489@utastro.UUCP>
Reply-To: israel@tove.UUCP (Bruce israel)
Distribution: net
Organization: U of Maryland, Laboratory for Parallel Computation, C.P., MD
Lines: 33

In article <489@utastro.UUCP> nather@utastro.UUCP (Ed Nather) writes:
>We have been running Unix 4.2bsd for a *long* time now, with very high
>load averages every day.  I guess it was inevitable, but strange effects
>on many working programs have been traced to a common cause:
>
>      /dev/null is full, and is overflowing!
>
>Anybody seen this problem before?  Can anyone help?

Yeah, we had it and discovered the solution quite by accident.  What
happened was that one of our staff people tried to empty the bit
bucket.  And how do you empty the bit bucket?  Obviously into /dev/null.
So he did a 'cat /dev/null >>/dev/null'.  After he cleaned up the mess
(it went all over the floor), he found that it hadn't solved it yet,
and the bit bucket was now twice as full.  Being a pretty ignorant staff
member, he kept on trying it.  Eventually the weight of all the one bits
was such that /dev/null collapsed onto itself and became a small black
hole.  We spoke to the physics dept. and found that it requires 2.73 x
10**42 one bits to reach critical mass (of course as everyone knows, zero
bits don't weigh anything and so are absolutely worthless in this procedure).
Anyway, now our /dev/null is a black hole and we never need to empty it.

Currently, one of our hackers is building a device driver so that we can
run mknod to make a black hole immediately without having to go thru the
above, long, very messy procedure.  Our first test of it will be to
install it in place of /usr/spool/news/net/flame/.

If anyone else is having this problem, I'd be glad to 'uucp' our /dev/null
file to your system.
-- 

Bruce Israel   
seismo!umcp-cs!israel (Usenet)    israel@Maryland (Arpanet)