[comp.protocols.nfs] handling RARP troubles in PC-NFS

milne@ICS.UCI.EDU (Alastair Milne ) (02/02/90)

   Since our installation had been using RARP for booting PC-NFS for quite 
   a while, and eventually had to give it up, I thought it might be 
   of interest for others to hear what happened.

   When we first installed PC-NFS (v 3.0) we tried to have it use RARP, and we
   thought we were reasonably successful (once we got past problems with
   interrupt numbers).  The PC's would boot to the net, though they were
   pretty laggardly about it.  I believe the Sun's that could volunteer
   PC-NFSD service are all Sun 3's, so perhaps we were simply putting up 
   with not having Sun 4's.

   However, we found that the frequency of successful booting was going down.
   We were used to PC's hanging during booting from time to time, and having
   to be re-booted after 3 minutes or so of apparently waiting for something
   from the net which was never coming -- but it was getting to be that you
   would expect to have to boot a machine 2 or 3 times before the net was
   properly up.  Unfortunately, the details surrounding this are a bit hard 
   to keep straight, so the description is not as clear as I'd like.  We
   updated to 3.0.1 at one point, and I thought performance was improving,
   but ultimately the failure frequency just kept going up.  We were also
   moved off the backbone of the department's net to a subnet from our PC-NFSD
   server.  This should have improved performance, since the backbone was
   getting busier and busier, but it didn't solve anything.

   Finally our network services manager told me he'd been looking at how 
   PC-NFS boots and what PC-NFSD does for it, and told me that he thought
   the RARP request was asking, not for a YP server directly, but for 
   any respondent to tell it what that machine's YP server was, and the 
   booting PC would then try to use that server too.  Since a fair number of
   YP servers are available which don't serve our subnet, but could well serve
   any recipient of our RARP broadcast packets, it was quite likely we'd wind
   up requesting a server that couldn't service us.

   So he suggested discarding RARP, and giving each PC its own little host
   table instead.  So I did that, including only the PC's own name and IP
   number, and those of a couple of available servers -- including the PC-NFSD
   server, of course.  That lets the PC boot to the net just about as fast as
   the server will respond.  Then I issue a NET YPDOMAIN xxxx and a 
   NET YPSET *, and we have back Yellow Pages.

   Note that, as far as I know, all information but what's in HOSTS is still
   obtained from the net, including the subnet mask.

   This arrangement has been doing us fine.  All the PC's boot the net much
   faster than they ever did before.  My only regret is that if there are any
   more addressing changes, I'll have to revise at least 8 host tables by
   hand.  Our site has been through at least 3 address changes so far, and
   RARP has saved me from having to lift a finger for them.  But I can't deny
   booting performance without it has been much better.

   Anybody else had this kind of luck with RARP?


   Alastair Milne,
   U.Calif. Irvine