Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!ames!ll-xn!mit-eddie!uw-beaver!cornell!batcomputer!itsgw!steinmetz!uunet!mcvax!cernvax!lid From: lid@cernvax.UUCP (lid) Newsgroups: comp.sys.apollo Subject: Re: Rwhod problems Message-ID: <727@cernvax.UUCP> Date: 22 Jun 88 09:18:39 GMT References: <8806162110.AA12175@caesar.cs.mtsu.edu> Reply-To: lid@cernvax.UUCP () Organization: CERN European Laboratory for Particle Physics, CH-1211 Geneva, Switzerland Lines: 20 Yes, I have an idea, get the tcp3.1, it is MUCH better ! Why ? 1) tcp_server initializes and when done forks itself, so you can start it up with 'cps -w' and the following tcp servers (ftp, inetd, whatever) will work correctly. 2) is much more robust, I run (via cron) a script every 1/2 hour to restart tcp servers if any of them go crazy: with previou TCPs they were restarted on average once a day (I was running tcp+ftp+ syslog+sendmail), now it seems to stay up almost forever. 3) tcp3.0 has a bug that has to do with subnets (e.g.: say your address is 100.0.1.55, you can contact all hosts 100.0.1.??, no problem, but you cannot reach, say, host 100.0.2.45: tcp thinks is a different network and hasn't got a gateway to it ! check with /usr/ucb/netstat -r and you'll see, it won't show routes to 100.0.0.0, but routes to 100.0.1.0). Last remark: 'cps -w' does not work in the server_process_manager startup script, oh why Apollo introduces these annoying bugs ? Are we supposed to run tcp gateways only on user nodes ? Achille Petrilli, Cray and PWS Operations