FTP problems [Was: Symbolics terminal emulator works lousy]

["Eric S. Crawley" <Crawley@alderaan.scrc.symbolics.com>]

    Date: Fri, 18 Dec 87 09:38:42 +0100
    From: mcvax!jungfrau!ceb@uunet.UU.NET

    I have run into this bug using the following operating systems:
    1. Ultrix 1.2
    2. Sun OS 3.2
    3. Bell Labs Version 9 Unix 
    4. VMS 4.5
    Modulo the normal operating system differences, it seems to appear
    consistently.  To be quite accurate,
    a. When you reboot a Lispm and there are ftp sessions or telnet
       sessions open, they get left around
    b. Under certain circumstances, when you type c-m-Abort and you are
       reading a file, you also get the associated ftp session added to
       your pile of junk deamons.
    Now, the system managers of all these systems do eat lunch together,
    so there is a possibility that they have developed some local bizarre 
    variation of the TCP parameters, but I somehow doubt it.

    Other similar services between other types of machines on the net
    clean up after themselves with no problems, and under similar
    conditions. 

To answer A:  Do you logout before you reboot the lispm?  Logout causes
all existing connections to be reset.  If you just do a :Halt Machine,
the connections are not closed.  Remember that you can continue after
halting the machine so it would not be correct for :Halt Machine to
clear out the connections.  That is why the functionality is in the
:Logout and :Reset Network commands.

Connections are kept to servers so they can be reused for subsequent file
transfers.  The file connection scavanger will remove a connection to a
server after the connection has been idle longer than
file-control-lifetime of the server.  The default for
file-control-lifetime is one hour.  You can make this smaller by putting
a value in the namespace object for the server in the
file-control-lifetime field.  The current units for that field are 60ths
of a second.  A value of 54000 will cause the connection to be closed
after 15 minutes.

I just tried to duplicate the behavior of your point B and the only
connection left around in most cases was the control connection that
would be scavanged normally.  Now, I have seen connections get left
around in semi-closed states that don't get scavanged but I have not
been able to find the cause or make it happen with any regularity.  I
have never seen the problem occur when the file-control-lifetime of the
server was set to a value less than 15 minutes.