"www/tcp server failing (looping), service terminated"

luotonen@ptsun00.cern.ch (Ari Luotonen)

Mail folder: WWW Talk Oct 93-present
Next message: Terry Allen: "URN ---> URN"
Previous message: Christopher J. McRae: "ANNOUNCEMENT: Second SIG-Web meeting (Palo Alto, CA)"

Date: Thu, 28 Oct 93 14:04:18 +0100
From: luotonen@ptsun00.cern.ch (Ari Luotonen)
Message-id: <9310281304.AA16669@ptsun00.cern.ch>
To: www-talk@nxoc01.cern.ch
Subject: "www/tcp server failing (looping), service terminated"


We have recovered the reason for info.cern.ch so often "refusing
connection".  This is not only our problem but could happen to anybody
having any daemon started very often by inetd on a Sun or a NeXT (and
maybe even on other platforms).

There is a hard-coded value in inetd allowing max 40 connections
from the same host per minute.  Fast machines can make more requests
(close to 60 per minute), which causes the inet daemon to detect this
as looping, and terminates the service for a while.

I don't yet know what browser or other program is used to cause this,
but it is obviously something that retrieves all the documents pointed
to by a given document.

In Solaris this is fixed, for SunOS 4.1.* there is a fix, e.g.

	ftp://qiclab.scn.rain.com/pub/sunos-patches/100178-08

which introduces -r command line option for inetd, e.g.:

	inetd -r 65 60

allows 65 connection per 60 seconds.

www1.cern.ch is a Sun, I have fixed it, but info.cern.ch is a NeXT --
we'll keep looking for a fix for it (anybody out there know anything
that would help??).

--
Ari Luotonen		 |
World-Wide Web Project	 |
CERN			 | phone: +41 22 767 8583
CH - 1211 Geneve 23	 | email: luotonen@dxcern.cern.ch