home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!dtix!darwin.sura.net!udel!princeton!siemens!aad
- From: aad@siemens.com (Anthony Datri)
- Newsgroups: comp.sys.sun.admin
- Subject: Re: default route for diskless clients
- Message-ID: <1992Dec28.151142.2167@siemens.com>
- Date: 28 Dec 92 15:11:42 GMT
- References: <1992Dec28.044346.4184@tc.cornell.edu>
- Sender: news@siemens.com (NeTnEwS)
- Organization: Siemens Corporate Research, Princeton (Plainsboro), NJ
- Lines: 72
- Nntp-Posting-Host: lovecraft
-
- >The default route for a diskless client is provided by rpc.bootparamd,
- >and it chooses the first entry in the kernel's network routing table
- >which is on the same subnet as the client.
-
- I had several hundred diskless machines (against my wishes) in a previous
- life, and what you describe sounds pretty much like what I saw.
-
- >This can lead to some
- >silly choices, such as one end of a PPP link, or a gateway to an
- >optical network. Both of those have happened, with no contortions on
- >my part.
-
- I didn't have those sorts of problems because our net was a little different
- from yours. Each subnet was connected to one router, and we didn't have
- PPP or SLIP links to worry about, and if we did, they'd have been on the
- router, not on hosts. And the whole net was optical 8^)
-
- >It might seem irrelevant which default route is used
-
- I was wondering that, since I always had rc.whatever explicitly set one through
- the router anyway. We made sure that the router was always host 254 on each
- subnet, which made it easy.
-
- >However, if the client's /usr lives on another subnet, there is a
- >chicken-and-egg problem.
-
- Wow. I have to admit that all of mine were on the same net. Is it safe to
- assume that there's some reason you can't move the client to the other net?
-
- >The 4.1.3 /etc/rc.boot mentions a statically linked route command, but
- >I think that's just there to torment people like me. Providing a
- >statically linked route command as part of the SunOS distribution
- >would be almost as useful as providing a dynamically linked mount
- >command.
-
- Weeeeeeel, ftp to your fav-o-rite BSD source site and get yerself the
- sources. You can find them on ftp.uu.net under systems/unix/bsd-sources,
- for example.
-
- I can't claim that I've built from these sources, but in said previous life
- I *did* build mount from that vendor's sources to try to get a binary that
- would reliably mount a pcfs filesystem. I gave up on that when I found mtools,
- though.
-
- >I could put libc.so.* into the client root partition, but that's gross.
-
- True enough, but you could hard-link the copies of it together for all of
- the clients like I did with kernels.
-
- >I could decide against mounting /usr from another subnet, but that's
- >not what I want.
-
- The typical argument for that decision would be speed, but if you've got a
- nifty router like a Cisco AGS+ then it's not so much of an issue.
-
- >What are my other options? Is there some way to tell bootparamd what
- >route to use for each client,
-
- Man, I miss not being at a vendor any more 8^(.
-
- >client)? Is there some way to rig the routing tables so that the
- >default route comes first?
-
- Well, you're not running routed, are you? If so, don't. Have the "server"
- set the default route first upon startup and then explicitly set any
- non-default routes with /usr/etc/route. I would think that this would
- ensure the desired order in the kernel table, not to mention making netstat -r
- far more pleasant.
-
- --
-
- ======================================================================8--<
-