[6304] in www-talk@info.cern.ch
Re: Network Abuse by Netscape? -- Was: Mosaic replacements, etc...
daemon@ATHENA.MIT.EDU (Simon E Spero)
Tue Oct 25 09:01:28 1994
Date: Tue, 25 Oct 1994 13:58:54 +0100
Errors-To: postmaster@www0.cern.ch
Errors-To: postmaster@www0.cern.ch
Reply-To: ses@tipper.oit.unc.edu
From: Simon E Spero <ses@tipper.oit.unc.edu>
To: Multiple recipients of list <www-talk@www0.cern.ch>
In theory, multiple simultaneous connections are a good idea; high-performance
servers are designed to perform well even when dealing with large numbers of
simultaneous connections. However, there are some problems with this approach.
The first problem occurs when networks start to become congested. TCPs congestion
control algorithms will will limit the damage, but because of the large number
of independently flow-controlled connections, will tend to back off far more
than would otherwise occur, giving much reduced throughput.
Another problem which is not mccs fault is caused by a bug in the TCP
implementation used by most commercial SYS V.4 releases. This bug causes
sockets to get stuck in close states when large numbers of connections
come in. This is the root of the problem that brought the NCSA ftp server
offline a while back, and is still a major headache on sunsite (there are
fixes in the offing from most vendors, but it looks like a hard problem).
At the moment it would probably be a good idea to disable the parallel gets
from netscape. If people are having real problems with servers going down, I
can also post some WEBTHRA patches for NCSA and CERN to send error messages
informing people about the o/s bugs when accessed by the problematic version
of netscape. This isn't really a good solution to the problem (it's the O/S
vendors fault, not MCC), and IBM is about to release a browser that does the
same thing, which would need yet another fix, but it should serve for
damage limitation.
Simon
