NetNews Usenet Archive 1992 #23/NN_1992_23.iso/spool/vmsnet/networks/tcpip/multinet/2219

Internet Message Format | 1992-10-09 | 1.5 KB

Path: sparky!uunet!ssdc!eseeger
From: [e-mail] (Eric Seeger)
Newsgroups: vmsnet.networks.tcp-ip.multinet
Subject: set socket options
Message-ID: <[e-mail]>
Date: 8 Oct 92 20:09:48 GMT
Organization: Sterling Software IMD, Vienna, Va.
Lines: 20

I have a simple question/observation concerning the socket option
SO_KEEPALIVE.  How can the documentation state that "Keepalives
are a questionable use of the network in that they cause idle
connections to add network traffic by constantly probing their
peer.".  When if fact by setting up a connection with this option
and caputuring what is put out on the network using TCPDUMP, we found
that it was 2 hours (TCPTV_KEEP_IDLE) before it sent out the first
probe packet to the peer.  Then every 75 seconds (TCPTV_KEEP_INTVL)
TGV will probe the peer up to a max count of 8 (TCPTV_KEEPCNT) before
he closes the connection, which amounts to a 2hr 10min timeout with
8 ack packets put out on the network over a 10 min. period.
That to me certainly does not seem to be a questionable use of the
network or any great amount of additional activity to detect idle
connections and clean them up. Is there something that the capture
of TCPDUMP is missing??

Additional question:
    Can the values of these symbols be changed so we can set the
timeout to what we might like before TGV closes idle sockets?

Eric J. Seeger                Phone:    [phone]
Sterling Software Inc.             UUCP:  uunet!ssdc!eseeger
8000 Towers Crescent Dr.         MAIL:  [e-mail]
Vienna, VA. 22182              FAX:  [phone]