Original Question
-----------------
>I have a Sun 670MP running SunOS 4.1.3 with many common patches with two
>ethernet interfaces. We are currently using 2 class c addresses and will be
>upgrading to a single class b address (ie subnetted block). I would like
>to put both ethernet interfaces on the same cable. One for class c and
>one for class b. This will allow me to slowly convert all our machines
>from class c to class b on the same physical network cable. I expect this to
>last for a few weeks at least.
>
>Is there any trouble doing this? Are any kernel patches required?
>Are there any kernel configuration changes required?
>
The first one to come up with the solution was mharris@jpmorgan.com
Solution
--------
Yes it can be done. It is okay to have two ethernet interfaces on the
same network. The two interfaces would each have their own IP addresses
one for each of the two networks (ie class C & class B).
If you do nothing else other than setting the IPs, moving the cable and
rebooting then it will work fairly well except... ;-)
The problem comes in with IP broadcasts. What happens is, when a broadcast
packet is set out it is picked up by the 1st ethernet interface and
rebroadcast out on the 2nd ethernet interface and the picked again by the 1st
ethernet interface and rebroadcast out on the 2nd ethernet interface ...
So a single IP broadcast packet is rebroadcast maybe 256 times. Various
people who responded called this a 'broadcast storm'. The result being that
the network seems to hang for a second or two, every so often. But everything
works the pretty much okay. If you are doing system performance monitoring
you will see this broadcast storm as a high 'socket overflow' number.
You can tell if you are having a broadcast storm by running this command
as root on another machine.
# etherfind -ip -and -broadcast
Normally you will get a few entries from various machines every few seconds.
During a broadcast storm you will get dozens of duplicates almost continuously.
One possibility is:
- turn off the routing 'in.routed -q', turn off IP forwarding and use static
routing.
The better choice is:
- turn off the IP Directed broadcasts, in the kernel, making it impossible for
it to forward broadcast packets. You can either use adb on /vmunix and/or
/dev/mem (so it takes effect immediately, without rebooting), or I think you
can edit something in a config file (I prefer adb, since it's faster).
This should do it, both permanently, and immediately:
# adb -k -w /vmunix /dev/mem
ip_dirbroadcast/D (to display current value)
ip_dirbroadcast/W0 (to turn off IP directed broadcasts on running kernel)
ip_dirbroadcast?W(-1) (to turn it off permanently in the /vmunix image)
<ctrl d>
# adb -k -w /vmunix /dev/mem
ip_dirbroadcast/D (to display current value)
ip_dirbroadcast/W1 (to turn on IP directed broadcasts on running kernel)
ip_dirbroadcast?W1 (to turn it on permanently in the /vmunix image)
<ctrl d>
The winning solution came from Sun Microsystems.
Special thanks to all those of you who replied especially mharris@jpmorgan.com.
Below are the original replies.
-------------------------------------------------------------------------
>From mharris@jpmorgan.com Mon Nov 29 09:07:02 1993
Conceptually, no problem. In fact, this is a good way to transition.
Now the bad news: Suns (and many other Unix systems) program the
hardware ethernet address of their ethernet interfaces based on
hostid. While this is no problem on different physical segments,
you have an arp problem when both if's are connected to the same
segment and have the same network address.
The workaround: ifconfig has an "ether" address family option
which allows you to explicitly set the network interface addr
so you can override the default setting on one interface.
See man 8 ifconfig.
- Assign a different ethernet address to the 2nd ethernet interface.
# vi /etc/rc.boot
- find the lines:
interface_names="`shcat /etc/hostname.* 2>/dev/null`"
if test -n "$interface_names"
then
(
IFS="$IFS."
set `echo /etc/hostname\.*`
while test $# -ge 2
do
shift
if [ "$1" != "xx0" ]; then
ifconfig $1 "`shcat /etc/hostname\.$1`" netmask + \
-trailers up
fi
shift
done
)
fi
- right after this, assign a different ethernet address to the 2nd
ethernet interface by adding the following line:
ifconfig le1 ether x:x:x:x:x:x where x is a hexadecimal digit
it must be one that is totally
unique and unused on your network
- Reboot your machine and everything should be perfect.
>From dan@bellcore.com Sat Nov 27 00:38:47 1993
All the sun ethernet interfaces on a sun workstation use the same
ethernet address. It is apparently a complex function of the hostid.
If you want to use two interfaces on the same cable, you must convince
one of the ethernet interfaces to use a different ethernet address.
I have no idea how you might do this. Perhaps there is an undocumented
ifconfig option.
Dan Strick, aka dan@bellcore.com or bellcore!dan, (201)829-4624
>From miker@il.us.swissbank.com Sun Nov 28 21:02:57 1993
BSD kernels (like SunOS 4.x) don't permit two addresses on the same interface.
And you really don't want to combine two separate networks on one cable; what's
the utilization level going to look like??
>From @a.gec-epl.co.uk:steve@gec-epl.co.uk Mon Nov 29 04:00:27 1993
um, yes. suns don't get their ethernet addresses from their card, but from
their internal eprom, which means that the two cards will have the same
address. there are ways around this, but that basically involves conning
the cards, and you need special interface software for that...
steve
>From walters@prl.philips.co.uk Mon Nov 29 10:28:07 1993
I don't know whether it's true, but I've just come off an Advanced Sun Admin
course, and the guy who gave us the course said that ethernet controllers
within a Sun had the same ethernet address - he said this is okay since they
never appear on the same piece of wire, and the address is in the PROM.
>From poffen@San-Jose.ate.slb.com Mon Nov 29 10:46:48 1993
It *SHOULD* work, no guarantees though. When we had to do a similar
conversion, we used a Cisco router, which can have multiple nets on the
SAME interface, and do the routing as well (pretty neat).
It will increase the traffic substantially on the wire though, because of all
the routed traffic.
>From strombrg@pleiades.acs.uci.edu Mon Nov 29 13:29:41 1993
Here are some articles I saved on the subject.
I've heard bad things about trying to use two different netmasks -
should make an interesting summary. If you try it out, I'd really
benefit from hearing how it went, as much or more than hearing what
people suggest.
Path: news.service.uci.edu!usc!elroy.jpl.nasa.gov!ames!cronkite.cisco.com!tli
From: tli@cisco.com (Tony Li)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: multiple subnets on same physical wire
Date: 8 Nov 1993 00:58:59 GMT
Organization: cisco Systems, Inc., Menlo Park, Ca.
Lines: 43
Distribution: na
Message-ID: <2bk5kjINN5jb@cronkite.cisco.com>
References: <CFy4L2.Ipy@world.std.com>
NNTP-Posting-Host: lager.cisco.com
In article <CFy4L2.Ipy@world.std.com> dpi@world.std.com (Mike Bloom) writes:
A very brief question: is it legal/possible to have several subnets share
the same piece of physical thinnet cable? A test customer of mine had the
following arrangement, but I'm not sure that it was legal:
1. Two subnets (e.g., 16.122.144 and 16.122.128) in the same building, each
with many Unix workstation nodes.
Yes, this is legal.
2. Two routers were configured as well, namely, one routed from subnet 144
to 128 while the other routed from 128 to 144. The routers were also on
the same physical wire, I believe.
This is strange, to say the least.
3. I tried to ping one node on subnet 128 from another on subnet 144, but
this failed.
4. I tried to construct static routes from the workstation on 128 through
the 128-144 router, as well as one in the opposite direction, but the
attempt to establish a route from 144-128 just hung. The ping failed
too. The customer's system administrator felt that the routers had
all of the routine information they needed without the need for extra
static routes, but we tried it anyway.
Most of the routers that I know of would not require additional static
routes.
5. It may be possible that the node on subnet 128 did not have its IP
address correctly set in it, but this is simply speculation at the
moment. Clearly, this would explain the symptoms that we saw!
Any info on what we did wrong and what we should do next would be
appreciated.
Well, I would start by simplifying the problem and only using one router to
route in both directions. I would then verify that you really are on one
physical cable. You should do this by snooping the wire, NOT just by
tracing it physically. Then snoop to see that the two hosts are getting to
the router.
Tony
Path: news.service.uci.edu!usc!howland.reston.ans.net!europa.eng.gtefsd.com!uunet!ddsw1!not-for-mail
From: karl@Notwerk.mcs.com (Karl Denninger)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: multiple subnets on same physical wire
Date: 7 Nov 1993 22:51:38 -0600
Organization: MCSNet Ops, Chicago, IL
Lines: 53
Distribution: na
Message-ID: <2bkj8q$2uo@Notwerk.mcs.com>
References: <CFy4L2.Ipy@world.std.com> <2bk5kjINN5jb@cronkite.cisco.com>
NNTP-Posting-Host: cluster.mcs.com
In article <2bk5kjINN5jb@cronkite.cisco.com>, Tony Li <tli@cisco.com> wrote:
>In article <CFy4L2.Ipy@world.std.com> dpi@world.std.com (Mike Bloom) writes:
>
> A very brief question: is it legal/possible to have several subnets share
> the same piece of physical thinnet cable? A test customer of mine had the
> following arrangement, but I'm not sure that it was legal:
>
> 1. Two subnets (e.g., 16.122.144 and 16.122.128) in the same building, each
> with many Unix workstation nodes.
>
>Yes, this is legal.
Legal but <highly> dangerous. Sun machines, in particular, will fail in
nasty ways with arp and broadcast storms if you are not <extremely> careful
doing this. The result will be cable meltdown, and it is not pretty.
Sun gear can be made "safe" for this, but it requires non-standard kernel
config options (at least with SunOS 4.x; no idea about Solaris 2.x) I've
had to do this on more than one occasion, usually for short term changes of
configuration where you had to get new things up before tearing down old
(like infrastructure -- cable plant, etc).
One thing to remember -- it is <absolutely essential> that if a SunOS machine
has two interfaces on the same wire, with different network numbers (as
defined by the netmask), that they be configured NOT to forward packets.
Note that this is exactly the opposite of the default config for a kernel
which has two interfaces active! It also means that a machine such as this
cannot route between the subnets for you.
> 2. Two routers were configured as well, namely, one routed from subnet 144
> to 128 while the other routed from 128 to 144. The routers were also on
> the same physical wire, I believe.
>
>This is strange, to say the least.
Also legal, but again, watch out for arp and broadcast storms! It is
possible to get redirect loopbacks in this situation if there is any
broadcast traffic on the wire (and there always is -- arps at minimum).
>Well, I would start by simplifying the problem and only using one router to
>route in both directions. I would then verify that you really are on one
>physical cable. You should do this by snooping the wire, NOT just by
>tracing it physically. Then snoop to see that the two hosts are getting to
>the router.
>
>Tony
Good advice. Grab a Sniffer and have at it.
-- Karl Denninger (karl@MCS.COM) | MCSNet - First Interactive Internet and Modem: [+1 312 248-0900] | Clarinet feed in Chicago. Send email to Voice/FAX: [+1 312 248-8649] | "info@mcs.com" for more information. Path: news.service.uci.edu!usc!cs.utexas.edu!uunet!psinntp!crynwr!nelson From: nelson@crynwr.com (Russell Nelson) Newsgroups: comp.protocols.tcp-ip Subject: multiple subnets on same physical wire Distribution: world Message-ID: <752903547snx@crynwr.com> References: <1993Nov8.171917.28674@ultb.isc.rit.edu> Date: Wed, 10 Nov 93 03:52:27 GMT Organization: Crynwr Software Lines: 18In article <2bk5kjINN5jb@cronkite.cisco.com>, Tony Li <tli@cisco.com> wrote:
In article <CFy4L2.Ipy@world.std.com> dpi@world.std.com (Mike Bloom) writes:
A very brief question: is it legal/possible to have several subnets share >> the same piece of physical thinnet cable? A test customer of mine had the following arrangement, but I'm not sure that it was legal:
1. Two subnets (e.g., 16.122.144 and 16.122.128) in the same building, each with many Unix workstation nodes.
Yes, this is legal.
Legal yes, but everyone must be running a TCP/IP package that knows how to deal with ICMP redirects and/or RIP. Otherwise, you'll always be transmitting *to* your default route(r), which will send back an ICMP redirect (which you'll ignore), and re-send the packet to the right host. Triple the outgoing traffic.
-russ <nelson@crynwr.com> ftp.msen.com:pub/vendor/crynwr/crynwr.wav Crynwr Software | Crynwr Software sells packet driver support. 11 Grant St. | 315-268-1925 (-9201 FAX) | Quakers do it in the light Potsdam, NY 13676 | LPF member - ask me about the harm software patents do. Path: news.service.uci.edu!usc!howland.reston.ans.net!usenet.ins.cwru.edu!magnus.acs.ohio-state.edu!dunedin!maf From: maf@dunedin.acs.ohio-state.edu (Mark Fullmer) Newsgroups: comp.protocols.tcp-ip Subject: Re: multiple subnets on same physical wire Date: 10 Nov 1993 17:33:55 GMT Organization: Ohio State University Academic Computing Services Lines: 34 Distribution: world Message-ID: <2br8m3$on9@charm.magnus.acs.ohio-state.edu> References: <1993Nov8.171917.28674@ultb.isc.rit.edu> <752903547snx@crynwr.com> NNTP-Posting-Host: dunedin.acs.ohio-state.edu
In article <752903547snx@crynwr.com> nelson@crynwr.com (Russell Nelson) writes: >In article <2bk5kjINN5jb@cronkite.cisco.com>, Tony Li <tli@cisco.com> wrote: > > In article <CFy4L2.Ipy@world.std.com> dpi@world.std.com (Mike Bloom) writes: > > A very brief question: is it legal/possible to have several subnets share >> the same piece of physical thinnet cable? A test customer of mine had the > following arrangement, but I'm not sure that it was legal: > > 1. Two subnets (e.g., 16.122.144 and 16.122.128) in the same building, each > with many Unix workstation nodes. > > Yes, this is legal. > >Legal yes, but everyone must be running a TCP/IP package that knows >how to deal with ICMP redirects and/or RIP. Otherwise, you'll always >be transmitting *to* your default route(r), which will send back an >ICMP redirect (which you'll ignore), and re-send the packet to the >right host. Triple the outgoing traffic.
Actually it's worse than this since the router can't send ICMP redirects. It would need to be an ARP redirect, which ofcourse doesn't exist.
Using RIP sorta works, but the clients need to listen to the RIP advertisements from the router, then install those routes as local routes. I did this once - the problem is it won't work for DOS machines, that can't run routed.
It's easier in the end to just change the subnet mask to have the hosts think they can talk to the entire network, then have the router do proxy arp for the subnets on the 'other' ports.
-- mark maf+@osu.edu Newsgroups: comp.protocols.tcp-ip Path: news.service.uci.edu!paris.ics.uci.edu!csulb.edu!library.ucla.edu!europa.eng.gtefsd.com!howland.reston.ans.net!math.ohio-state.edu!caen!crl.dec.com!crl.dec.com!nntpd.lkg.dec.com!ryn.mro4.dec.com!janix.pcs.dec.com!janix.pcs.dec.com!pcsrcu!tartufo!me From: me@tartufo.pcs.com (Michael Elbel) Subject: Re: multiple subnets on same physical wire Reply-To: me%dude.pcs.dec.com@inet-gw-2.pa.dec.com Organization: PCS Computer Systeme GmbH Date: Fri, 12 Nov 1993 09:16:47 GMT Message-ID: <me.753095807@tartufo> References: <1993Nov8.171917.28674@ultb.isc.rit.edu> <752903547snx@crynwr.com> <2br8m3$on9@charm.magnus.acs.ohio-state.edu> Sender: @pcsrcu.pcs.com Lines: 47
In <2br8m3$on9@charm.magnus.acs.ohio-state.edu> maf@dunedin.acs.ohio-state.edu (Mark Fullmer) writes:
>>Legal yes, but everyone must be running a TCP/IP package that knows >>how to deal with ICMP redirects and/or RIP. Otherwise, you'll always >>be transmitting *to* your default route(r), which will send back an >>ICMP redirect (which you'll ignore), and re-send the packet to the >>right host. Triple the outgoing traffic.
>Actually it's worse than this since the router can't send ICMP redirects. >It would need to be an ARP redirect, which ofcourse doesn't exist.
>Using RIP sorta works, but the clients need to listen to the RIP >advertisements from the router, then install those routes as local routes. >I did this once - the problem is it won't work for DOS machines, that >can't run routed.
>It's easier in the end to just change the subnet mask to have the hosts >think they can talk to the entire network, then have the router do proxy >arp for the subnets on the 'other' ports.
Maybe I'm not understanding the whole point, but what's wrong with just using static routes that point to the respective other net? I agree, you'll have to set them up, but so you have to do with the actual IP-Address and net mask.
We are running two different subnets in this building on the same wire (16.186.144.0 and 16.186.160.0). We have a router with addresses in both nets that runs rip (its actual job is to route to the rest of our net). Without doing anything special, just using routed, machines on both subnets can talk together through the router. This doesn't make much sense, however, since the packets get put onto the same wire again, effectively doubling the net load for communication between the nets.
All we had to do to have the net run properly was add static routes to the "other" net on all the workstations. E.g.
/etc/route add net 16.186.160 tartufo 0
on my machine tartufo (16.186.144.131)
Could somebody please enlighten me, if we actually have a possible problem with this solution?
Michael Michael Elbel, Digital-PCS GmbH, Muenchen, Germany - me@dude.pcs.dec.com Fermentation fault (coors dumped) Path: news.service.uci.edu!paris.ics.uci.edu!csulb.edu!library.ucla.edu!agate!usenet.ins.cwru.edu!magnus.acs.ohio-state.edu!dunedin!maf From: maf@dunedin.acs.ohio-state.edu (Mark Fullmer) Newsgroups: comp.protocols.tcp-ip Subject: Re: multiple subnets on same physical wire Date: 12 Nov 1993 16:48:08 GMT Organization: Ohio State University Academic Computing Services Lines: 40 Message-ID: <2c0eo8$20f@charm.magnus.acs.ohio-state.edu> References: <752903547snx@crynwr.com> <2br8m3$on9@charm.magnus.acs.ohio-state.edu> <me.753095807@tartufo> NNTP-Posting-Host: dunedin.acs.ohio-state.edu
In article <me.753095807@tartufo> me%dude.pcs.dec.com@inet-gw-2.pa.dec.com writes: >In <2br8m3$on9@charm.magnus.acs.ohio-state.edu> maf@dunedin.acs.ohio-state.edu (Mark Fullmer) writes: > >>It's easier in the end to just change the subnet mask to have the hosts >>think they can talk to the entire network, then have the router do proxy >>arp for the subnets on the 'other' ports. > >Maybe I'm not understanding the whole point, but what's wrong with >just using static routes that point to the respective other net? >I agree, you'll have to set them up, but so you have to do with the >actual IP-Address and net mask.
That's fine until you are forced to support IP stacks that only know about a default router - ie. many DOS applications.
>We are running two different subnets in this building on the same wire >(16.186.144.0 and 16.186.160.0). We have a router with addresses in >both nets that runs rip (its actual job is to route to the rest of our >net). Without doing anything special, just using routed, machines on >both subnets can talk together through the router. This doesn't make >much sense, however, since the packets get put onto the same wire >again, effectively doubling the net load for communication between >the nets. > >All we had to do to have the net run properly was add static routes to >the "other" net on all the workstations. E.g. > > /etc/route add net 16.186.160 tartufo 0 > >on my machine tartufo (16.186.144.131) > >Could somebody please enlighten me, if we actually have a possible >problem with this solution?
Your setup is correct.
-- mark maf+@osu.edu
Path: news.service.uci.edu!usc!elroy.jpl.nasa.gov!ames!cronkite.cisco.com!tli From: tli@cisco.com (Tony Li) Newsgroups: comp.protocols.tcp-ip Subject: Re: multiple subnets on same physical wire Date: 10 Nov 1993 21:33:12 GMT Organization: cisco Systems, Inc., Menlo Park, Ca. Lines: 13 Distribution: world Message-ID: <2brmmpINN55n@cronkite.cisco.com> References: <1993Nov8.171917.28674@ultb.isc.rit.edu> <752903547snx@crynwr.com> NNTP-Posting-Host: lager.cisco.com
In article <752903547snx@crynwr.com> nelson@crynwr.com (Russell Nelson) writes:
Legal yes, but everyone must be running a TCP/IP package that knows how to deal with ICMP redirects and/or RIP. Otherwise, you'll always be transmitting *to* your default route(r), which will send back an ICMP redirect (which you'll ignore), and re-send the packet to the right host. Triple the outgoing traffic.
Intelligent (and correctly configured) routers will notice that the source and the next hop are not on the same subnet, and will not send a redirect.
Tony
Newsgroups: comp.protocols.tcp-ip Path: news.service.uci.edu!usc!howland.reston.ans.net!agate!library.ucla.edu!nntp.ucsb.edu!fennel.acc.com!art From: art@acc.com (Art Berggreen) Subject: Re: multiple subnets on same physical wire Message-ID: <1993Nov11.165705.274@acc.com> Organization: ACC, Advanced Computer Communications References: <1993Nov8.171917.28674@ultb.isc.rit.edu> <752903547snx@crynwr.com> Date: Thu, 11 Nov 1993 16:57:05 GMT Lines: 14
In article <752903547snx@crynwr.com> nelson@crynwr.com (Russell Nelson) writes: >Legal yes, but everyone must be running a TCP/IP package that knows >how to deal with ICMP redirects and/or RIP. Otherwise, you'll always >be transmitting *to* your default route(r), which will send back an >ICMP redirect (which you'll ignore), and re-send the packet to the >right host. Triple the outgoing traffic.
I disagree. Router Requirements states that a router should not issue a Redirect unless the source host is on the same (sub)net as the next-hop. RFC1122 states that a host should ignore a Redirect to a (sub)net different than the one it is connected to.
Art
Path: news.service.uci.edu!usc!howland.reston.ans.net!math.ohio-state.edu!sdd.hp.com!portal!mack!news From: ddean@minerva.rolm.com (Drew Dean) Newsgroups: comp.protocols.tcp-ip Subject: Re: multiple subnets on same physical wire Date: 8 Nov 1993 20:33:03 GMT Organization: ROLM - A Siemens Company Lines: 23 Distribution: world Message-ID: <2bmae0$94v@mack.eng.sc.rolm.com> References: <2bkj8q$2uo@Notwerk.mcs.com> Reply-To: ddean@minerva.rolm.com NNTP-Posting-Host: opus.eng.sc.rolm.com Summary: Use a scope
|In article 2uo@Notwerk.mcs.com, karl@Notwerk.mcs.com (Karl Denninger) writes: |In article <2bk5kjINN5jb@cronkite.cisco.com>, Tony Li <tli@cisco.com> wrote: |> You should do this by snooping the wire, NOT just by |>tracing it physically. Then snoop to see that the two hosts are getting to |>the router.
|>Tony
|Good advice. Grab a Sniffer and have at it.
|Karl Denninger (karl@MCS.COM) | MCSNet - First Interactive Internet and
Well, you weren't at Wesley Irish's talk last week at Stanford. Sniffers have a nasty way of dropping packets, especially after collisions and other abnormal events. What you really need is a good digital storage oscilloscope, and a little knowledge of Manchester encoding. BTW, it's fun !
-- Drew Dean ddean@robadome.com -- I don't speak for ROLM....
Path: news.service.uci.edu!usc!howland.reston.ans.net!spool.mu.edu!think.com!think.com!barmar From: barmar@think.com (Barry Margolin) Newsgroups: comp.protocols.tcp-ip Subject: Re: multiple subnets on same physical wire Date: 9 Nov 1993 18:26:30 -0500 Organization: Thinking Machines Corporation, Cambridge MA, USA Lines: 48 Sender: root@Think.COM Distribution: na Message-ID: <9311092326.AA00894@telecaster.think.com> References: <CFy4L2.Ipy@world.std.com>
In article <2bkj8q$2uo@Notwerk.mcs.com> karl@Notwerk.mcs.com (Karl Denninger) writes: >In article <2bk5kjINN5jb@cronkite.cisco.com>, Tony Li <tli@cisco.com> wrote: >>In article <CFy4L2.Ipy@world.std.com> dpi@world.std.com (Mike Bloom) writes: >> 1. Two subnets (e.g., 16.122.144 and 16.122.128) in the same building, each >> with many Unix workstation nodes. >>Yes, this is legal. >Legal but <highly> dangerous. Sun machines, in particular, will fail in >nasty ways with arp and broadcast storms if you are not <extremely> careful >doing this. The result will be cable meltdown, and it is not pretty.
I had this arrangement for a while, and the only problem the Suns had were that routed complained "Packet from unknown router" every time it saw the RIP broadcast from the router's "other" address. This was solved by putting an entry for the unknown router in the Suns' /etc/gateways files.
I also had to configure my cisco router to use a 255.255.255.255 broadcast address rather than <network>.<subnet>.255, since the latter didn't work when trying to broadcast to the secondary subnet.
>Sun gear can be made "safe" for this, but it requires non-standard kernel >config options (at least with SunOS 4.x; no idea about Solaris 2.x) I've >had to do this on more than one occasion, usually for short term changes of >configuration where you had to get new things up before tearing down old >(like infrastructure -- cable plant, etc).
I never made any kernel changes for this, and my Suns were fine. Are you talking about using the Sun as the router between the two subnets?
>> 2. Two routers were configured as well, namely, one routed from subnet 144 >> to 128 while the other routed from 128 to 144. The routers were also on >> the same physical wire, I believe. >> >>This is strange, to say the least. > >Also legal, but again, watch out for arp and broadcast storms! It is >possible to get redirect loopbacks in this situation if there is any >broadcast traffic on the wire (and there always is -- arps at minimum).
I don't see why ARP would have any particular problems. Hosts will only ARP for destinations they think are on the same subnet. When talking to hosts on the other subnet, they'll ARP for the gateway, and the gateway will ARP for the destination, and everything should work fine.
I suppose there could be problems if you have machines configured with the wrong subnet mask (e.g. using the class B mask rather than the site's subnet mask), so they ARP for machines on the other subnet. If the router also does proxy ARP then both the router and the real host will respond to the ARP query.
In message <9311261337.AA19314@zeus.dciem.dnd.ca> you write: >I have a Sun 670MP running SunOS 4.1.3 with many common patches with two >ethernet interfaces. We are currently using 2 class c addresses and will be >upgrading to a single class b address (ie subnetted block). I would like >to put both ethernet interfaces on the same cable. One for class c and >one for class b. This will allow me to slowly convert all our machines >from class c to class b on the same physical network cable. I expect this to >last for a few weeks at least. > >Is there any trouble doing this? Are any kernel patches required? >Are there any kernel configuration changes required? > >Thanks and yes I will summarize.
Dan Stromberg - OAC/DCS strombrg@uci.edu
>From dougj@iplab.health.ufl.edu Mon Nov 29 13:56:19 1993 To: rwolf@dretor.dciem.dnd.ca Subject: Re: 2 ethernet interfaces on same network cable Content-Length: 85 X-Lines: 5 Status: RO
Should not be a problem. You will just have to setup the routing correctly.
dougj
>From sun1!sun1.philips.com!dkin@philabs.philips.com Mon Nov 29 14:41:18 1993 To: rwolf@dretor.dciem.dnd.ca Subject: 2 ethernet interfaces on same network cable X-Sun-Charset: US-ASCII Content-Length: 691 X-Lines: 13 Status: RO
Be careful with two interfaces and the same network address. We have a SUN 670MP with two interfaces and class B addresses. Sun automatically assigns the same ethernet address to each ethernet interface installed in the system. This will cause a problem when both interfaces are configured with the same network address (ie. 128.158.x.x). I had to manually assign a different ethernet address to the second interface card using ifconfig. After doing this everything worked. I assigned a second hostname and IP address to the second interface card. Also, I use static routing to reach specific hosts via the second interface.
No patches or kernel changes were needed.
David Kindle
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:08:33 CDT