Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 25 Oct 1998 16:46:12 -0600
From:      "Jeffrey J. Mountin" <jeff-ml@mountin.net>
To:        <ratbert@phoenix.aye.net>, Leif Neland <root@swimsuit.internet.dk>
Cc:        freebsd-isp@FreeBSD.ORG
Subject:   Re: route changes erratically (routed)
Message-ID:  <3.0.3.32.19981025164612.00ff9974@207.227.119.2>
In-Reply-To: <Pine.BSF.3.96.981025135922.28835A-100000@phoenix.aye.net>
References:  <Pine.BSF.4.05.9810230024380.1409-100000@gina.swimsuit.internet.dk>

next in thread | previous in thread | raw e-mail | index | archive | help
At 02:11 PM 10/25/98 -0500, ratbert@phoenix.aye.net wrote:
>
>We've had some similar problems with our portmasters, OSPF and rip2  both
>seemed to be broken on them. We assigned an x.x.x.x/28 for the dialup
>lines and the portmasters ended up broadcasting themselves as a route
>to a x.x.x.x/28 and /29s, /30s, /31s and /32s within the /28. 
>Pretty much turned the routing tables of everything on our network to
>complete garbage.

RIPv2 does not exist and most likely never will in the COM/OS and OSPF is
definately not broken.

As for garbage, it could be cleaned up with better planning and OSPF beats
the hell out of plugging static routes.

>What we did to solve it was add a static route on our servers and other
>routers with the portmaster as a gateway to the dialup subnet assigned
>to it. Have the static dialup ip addresses be on the same network with
>the portmasters and servers and let the portmasters proxyarp for those
>ip addresses.

Proxyarp advocate eh?

Unless a server is a gateway there is no reason to run a routing daemon,
unless you don't want the router to be a hop, but if the addresses are not
in the same /24 they will be. YMMV, but for simplicity and pertinence to
the original post.

For PM2's it works best if you start with the 2nd /27, use OSPF, and set
the pool size to 32.  Bam, one route!

Should you have a slew of these you start the first on .2 (not .1) and use
pool-size=30 (it can only have 30, but still) and you get:

.2  /31
.4  /30
.8  /29
.16 /28

If you use .1 you get:

.1  /32
.2  /31
.4  /30
.8  /29
.16 /29
.24 /30
.28 /31
.30 /32

Fairly visual example of why one *should* use even boundaries.  Tends to
add a bit of clutter, as you know. ;)

Expand this to a fully populated /24 with 8 PM2's:

.2   /31 - pm1 (pool size=30)
.4   /30
.8   /29
.16  /28
.32  /27 - pm2 (pool size=32 ditto for pm3-7)
.64  /27 - pm3
.96  /27 - pm4
.128 /27 - pm5
.160 /27 - pm6
.192 /27 - pm7
.224 /28 - pm8 (pool size=32)
.240 /29
.248 /30
.252 /31

Gosh, only 14 routes and some few lines in the Cisco or did you really want
64 routes?  Didn't think so. ;)

>On Fri, 23 Oct 1998, Leif Neland wrote:
>
>> We have 2 portmasters (PM2), several servers, a cisco to the world, and a
>> firewall to the internal network.
>> 
>> The cisco is default gateway
>> 
>> The servers and portmasters are on one class C, the dialins are on another
>> class C.
>> 
>> Because some users have fixed ip, but can dial in on either of the
>> portmasters, I run routed on all servers, and the portmasters seem to
>> announce on which portmaster the customer is, so the route gets changed to
>> the right portmaster.

<no_flame>
Don't use routed, ever.
</no_flame>

You *could* use gated and OSPF for this but there is no reason with the
Cisco being the default gateway and for the size of your setup.  No issue
with the servers and PM IPs on one /24 and the dial-in IPs on another.
 
>> The traceroute should then go from server to pm1 or pm2 to customer.

Only if you really want it that way.  You either live with one more hop and
a slight increase to the latency or a more complex setup for a small gain.
With OSPF you could inject RIP, but again for a small gain.  I didn't
bother since *most* traffic from dial-up will go out the router anyways.
 
>> However, often the route changes so it goes
>> server->cisco->pm->client or
>> server->firewall->pm->client or even
>> server->cisco->(router at our uplink)->cisco->(router at our uplink) etc.

The first 2 are ok, but the 3rd?  What version of COMOS on the PM2's?

>> If I constantly pings the client, I gets pauses where the pings are lost.

How are your network collisions?  Sounds like it may be bad cabling since
at least some packets are making it.

>> What do I do wrong? Shouldn't I use routed on the servers, but only route
>> default gateway to the cisco, and let it handle the pm1/pm2 route changes?
>> Or should I have one server running routed? or gated? or what?

Use OSPF between the PM's and the Cisco, verify propagation, lose the
static routes, turn off routed on the servers, and all is well.

I've got templates for the PMs so it would be a matter of changing a few
things and pasting it in a terminal window.  Takes less than 10 minutes to
convert.  One hitch is I don't have access to a Cisco at the moment so
would need either access to one (for just one IP - access control is good
:) or some pointers for Cisco OSPF, since I'm a bit rusty.  Not something I
do often, after all once you set it...

cheers!


Jeff Mountin - Unix Systems TCP/IP networking
jeff@mountin.net

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-isp" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3.0.3.32.19981025164612.00ff9974>