headers
christian (private | 1 Sep 2005 01:12
Picon
Picon

[quagga-users 5254] Re: OSPF design question

whatever design you choose, you should take care that the bb area does
not get "divided" to 2 seperate networks.
typically a full mesh helps in archieving this, as mostly one link can
fail and still full reachability can be maintained.
from a view of flooding in a shared environment, ospf handles this quite
well, because of DR and BDR election.

I prefer definately the full or partial mesh over the shared one. shared
one mostly has a single link to the hub/switch and is a spof.

as for subnetting, choose a /30 netmask for each link. a /24 is a great
waste of resoucres.
/30 handles exact 2 hosts (your router-link-adresses).
like 192.168.10.0/30
192.168.10.0 network
192.168.10.1 router 1
192.168.10.2 router 2
192.168.10.3 broadcast-adr

to your answer - make 1 subnet per link. in summary in your case: 5

christian

recvfrom@... wrote:

> Hi!
>
> While this question isn't strictly quagga-related, I hope I'm not out
> of line
> asking this here...
(Continue reading)

Permalink | Reply | 5 comments |
headers
paul cunnane | 1 Sep 2005 01:20
Picon

[quagga-users 5255] Re: OSPF design question

> as for subnetting, choose a /30 netmask for each link. a /24 is a great
> waste of resoucres.
> /30 handles exact 2 hosts (your router-link-adresses).
> like 192.168.10.0/30
> 192.168.10.0 network
> 192.168.10.1 router 1
> 192.168.10.2 router 2
> 192.168.10.3 broadcast-adr

Is it possible to use /31 in a case like this, and avoid using the
network and broadcast addresses?

--

-- 
Paul
Permalink | Reply | 2 comments |
headers
recvfrom | 1 Sep 2005 01:32
Picon

[quagga-users 5256] Re: OSPF design question



On 8/31/05, christian (private) <chr.graf-Mmb7MZpHnFY@public.gmane.org> wrote:
whatever design you choose, you should take care that the bb area does
not get "divided" to 2 seperate networks.
typically a full mesh helps in archieving this, as mostly one link can
fail and still full reachability can be maintained.
from a view of flooding in a shared environment, ospf handles this quite
well, because of DR and BDR election.

Yes, but in a partial mesh, if I understand correctly, I must make sure
that the DR and BDR have full physical connectivity to each of the other
routers, correct?  Meaning, that they must be the routers in the two
corners that can reach each of the other three routers.  If this is correct,
it would not be possible to have a fifth router directly connected to either
the DR or the BDR, but not both.  I also don't know what would happen
if either the DR or BDR failed, forcing a re-election.  Hmmm, seems
like a full mesh is safer!??

I prefer definately the full or partial mesh over the shared one. shared
one mostly has a single link to the hub/switch and is a spof.

as for subnetting, choose a /30 netmask for each link. a /24 is a great
waste of resoucres.

Yes, exactly what I had in mind!

/30 handles exact 2 hosts (your router-link-adresses).
like 192.168.10.0/30
192.168.10.0 network
192.168.10.1 router 1
192.168.10.2 router 2
192.168.10.3 broadcast-adr

to your answer - make 1 subnet per link. in summary in your case: 5

Thanks for the quick reply!!!

-r


recvfrom-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org wrote:

> Hi!
>
> While this question isn't strictly quagga-related, I hope I'm not out
> of line
> asking this here...
>
> I'd like to modify my backbone area to create a partial mesh (all
> ethernet),
> as shown below:
>
> +--------+            +--------+
> |   R1   |-------------|   R2   |
> +--------+            +--------+
>      |        \               |
>      |          \             |
>      |            \           |
>      |              \         |
>      |                \       |
> +--------+            +--------+
> |   R4   |-------------|   R3   |
> +--------+            +--------+
>
> Right now R1-4 are in a single broadcast domain (but not with the
> topology shown above -- no loops, just a simple bus), using one /24.
> With the above topology, I assume I need to further partition the /24.
> My question... should there be four subnets or five (one per link)?
> Any great disadvantages to doing this, besides a slightly more
> complicated configuration?  I only see the upside.  TIA!!!
>
> -r
>
>------------------------------------------------------------------------
>
>_______________________________________________
>Quagga-users mailing list
>Quagga-users-UOy77sIEA+cAd7ICUelF/Q@public.gmane.org
>http://lists.quagga.net/mailman/listinfo/quagga-users
>
>

--
Christian Graf
Brombeerweg 2
37281 Wanfried
Mob. 0049.163.2474723




_______________________________________________
Quagga-users mailing list
Quagga-users@...
http://lists.quagga.net/mailman/listinfo/quagga-users
Permalink | Reply | 1 comment |
headers
recvfrom | 1 Sep 2005 01:37
Picon

[quagga-users 5257] Re: OSPF design question



On 8/31/05, paul cunnane <paul.cunnane-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
> as for subnetting, choose a /30 netmask for each link. a /24 is a great
> waste of resoucres.
> /30 handles exact 2 hosts (your router-link-adresses).
> like 192.168.10.0/30
> 192.168.10.0 network
> 192.168.10.1 router 1
> 192.168.10.2 router 2
> 192.168.10.3 broadcast-adr

Is it possible to use /31 in a case like this, and avoid using the
network and broadcast addresses?

AFAIK, there are implementations that allow you to use the network address,
but not the broadcast address, so unless I'm terribly mistaken, the answer is
'no'.

-r



_______________________________________________
Quagga-users mailing list
Quagga-users@...
http://lists.quagga.net/mailman/listinfo/quagga-users
Permalink | Reply | 0 comments |
headers
John Fraizer | 1 Sep 2005 03:15

[quagga-users 5258] Re: multihomed router and loopback as source

Rosario Pingaro wrote:
> my question was not about what ip use to peer. I have to use loopback 
> because I have more then one line for each upstream.
> 
> This is about bgp.
> 
> But then when I do something on the internet with the router, the 
> packets are originated from the interface ip instead the loopback.
> I would like to exit to internet with one ip for all providers, the 
> loopback's ip, not just about to open the bgp session.
> 
> thanks

Under Linux, traffic originating FROM the router itself is going to use 
the PRIMARY address of the interface the packet is leaving as the source 
address.  This is just how it works.

Why are you using your router to *originate* traffic?  Let it route.

John
Permalink | Reply | 0 comments |
headers
John Fraizer | 1 Sep 2005 03:24

[quagga-users 5259] Re: multihomed router and loopback as source

Rosario Pingaro wrote:
> "something" does mean that if:
> - doing a ping to an host, and using ethereal on that host, i see the 
> icmp request arriving some time form on interface's ip, othe time from 
> different interface's ip

OK.  This is probably because the routing table updated during your ping 
and the preferred exit interface (preferred route) to your ping target 
changed.

> - doing a traceroute, it doesn't work because the packets have different 
> sources;

You can specify the source interface with traceroute.

Usage: traceroute [-dFInrvx] [-g gateway] [-i iface] [-f first_ttl] [-m 
max_ttl] [ -p port] [-q nqueries] [-s src_addr] [-t tos] [-w waittime]
host [packetlen]

Example:  If you wanted to traceroute with a source address of 10.1.1.1, 
you would do this:

traceroute -s 10.1.1.1 [target]

If you wanted to traceroute and exit ONLY via eth1, you would do this:

traceroute -i eth1 [target]

> - i cant update the the system using apt-get or yum because I am not 
> able to estabilish a stable connection because the requests sent from 
> the router have different ip sources.
> 

Sure you can.

You can:

(1) Static route to the server you're trying to contact, thus forcing 
the route AND your source IP address.

...or...

(2) Shut down your redundant BGP sessions so you only have ONE route to 
the server you're trying to contact.  (Same effect as above.)

> Maybe it is a regular behavior and a workeround it is not related to 
> quagga, but I was only searching help.
> 

Routing-101 my friend.  Routing-101.

John
Permalink | Reply | 0 comments |
headers
John Fraizer | 1 Sep 2005 03:44

[quagga-users 5261] Re: OSPF design question

paul cunnane wrote:
>>as for subnetting, choose a /30 netmask for each link. a /24 is a great
>>waste of resoucres.
>>/30 handles exact 2 hosts (your router-link-adresses).
>>like 192.168.10.0/30
>>192.168.10.0 network
>>192.168.10.1 router 1
>>192.168.10.2 router 2
>>192.168.10.3 broadcast-adr
> 
> 
> Is it possible to use /31 in a case like this, and avoid using the
> network and broadcast addresses?
> 

Not in a broadcast network, no.  a /31 can be used for p-t-p networks 
but, you're asking for problems using it with a broadcast medium.

John
Permalink | Reply | 0 comments |
headers
John Fraizer | 1 Sep 2005 03:42

[quagga-users 5260] Re: OSPF design question

recvfrom@... wrote:
> Hi!
> 
> While this question isn't strictly quagga-related, I hope I'm not out of 
> line
> asking this here...
> 
> I'd like to modify my backbone area to create a partial mesh (all 
> ethernet),
> as shown below:
> 
> +--------+            +--------+
> |   R1   |-------------|   R2   |
> +--------+            +--------+
>      |        \               |
>      |          \             |
>      |            \           |
>      |              \         |
>      |                \       |
> +--------+            +--------+
> |   R4   |-------------|   R3   |
> +--------+            +--------+
> 
> Right now R1-4 are in a single broadcast domain (but not with the
> topology shown above -- no loops, just a simple bus), using one /24. 
> With the above topology, I assume I need to further partition the /24. 
> My question... should there be four subnets or five (one per link)? 
> Any great disadvantages to doing this, besides a slightly more
> complicated configuration?  I only see the upside.  TIA!!!
> 
> -r
> 

Split your /24 into 2 /25's.  Then, take one of the /25's and split it 
further into 4 /27's.

Take one of those /27's and split it into 8 /30's.

Now, assign each link in your mesh it's own /30.

If you picked a.b.c.224/27 to carve your 8 /30's out, you would end up with:

a.b.c.224/30
a.b.c.228/30
a.b.c.232/30
a.b.c.236/30
a.b.c.240/30
a.b.c.244/30
a.b.c.248/30
a.b.c.252/30

Now set your links up as follows:

R1-R2:
R1=a.b.c.225/30
R2=a.b.c.226/30

R2-R3:
R2=a.b.c.229/30
R3=a.b.c.230/30

R3-R4:
R3=a.b.c.233/30
R4=a.b.c.234/30

R4-R1:
R4=a.b.c.237/30
R5=a.b.c.238/30

R1-R3:
R1=a.b.c.241/30
R2=a.b.c.242/30

This will leave 3 /30's for further meshing in the future.

Your network will now consist of:

a.b.c.0/24:
----------
	a.b.c.0/25
		a.b.c.128/26
			a.b.c.192/27
				a.b.c.224/30
				a.b.c.228/30
				a.b.c.232/30
				a.b.c.236/30
				a.b.c.240/30
				a.b.c.244/30
				a.b.c.248/30
				a.b.c.252/30

You will still announce your /24 as a /24 to your provider but, you'll 
have to route it as above internally.

Hopefully, you've got the space left in one end of you're /24 to split 
off a /25 so you can carve it up.  If not, you're in for some real fun 
renumbering devices to get them to one end or the other.  I had to do 
exactly what I've described above for a client (who shall remain 
nameless to protect the clueless) and they had customers (not enough to 
fill a single /24) spread all the hell throughout a /22.  The had to 
renumber 26 machines just to clear out a /24 enough that we could carve 
out a /27 that had to be routed differently.

John Fraizer
EnterZone, Inc
Permalink | Reply | 0 comments |
headers
Gobbledegeek | 1 Sep 2005 10:07
Picon

[quagga-users 5262] Quagga + Xen virtualization anyone?

Hello
       Anyone attempted installing quagga with Xen? Has anyone been
able to make virtual routers talk ospf or bgp to each other before?
Will they share the same NIC? What about the mac address? I wish to
try thisnext week if possible, and thought some advice will be
welcome...
--

-- 
Nonchalantly yours
GobbledeGeek
[Every thing but Gobbledegook.. !!]
Permalink | Reply | 2 comments |
headers
christian (private | 1 Sep 2005 10:27
Picon
Picon

[quagga-users 5263] Re: OSPF design question

DR and BDR are only elected when using on multiaccess-networks like
ethernet.
Even if you do connect directly two ospf-routers together, one will act
as the DR and the other can act as BDR. (the DR is mandatory, the BDR is
optional)
The concept of DR and BDR is just for saving unnecassary LSA flooding.
If the DR fails, the BDR will take over the function of DR. OSPF is
using the hello-protocol to elect DR.
The election is based on the priority and if they are equal, the highest
router-id is used a tiebreaker.
Once the DR is elected, a newly installed router with an even HIGHER
priority will not take DR-ownership. OSPF is not deterministic here.

It should be absolutely no problem adding a 5th router, which is direct
connected to the BR as well the BDR into this scenario. The 5th router
is using seperate links and on those you get as well a DR and BDR election.
If you are unsure, please drop a short topology and we can discuss.

christian

recvfrom@... wrote:

>
>
> On 8/31/05, *christian (private)* <chr.graf@...
> <mailto:chr.graf@...>> wrote:
>
>     whatever design you choose, you should take care that the bb area does
>     not get "divided" to 2 seperate networks.
>     typically a full mesh helps in archieving this, as mostly one link can
>     fail and still full reachability can be maintained.
>     from a view of flooding in a shared environment, ospf handles this
>     quite
>     well, because of DR and BDR election.
>
>
> Yes, but in a partial mesh, if I understand correctly, I must make sure
> that the DR and BDR have full physical connectivity to each of the other
> routers, correct?  Meaning, that they must be the routers in the two
> corners that can reach each of the other three routers.  If this is
> correct,
> it would not be possible to have a fifth router directly connected to
> either
> the DR or the BDR, but not both.  I also don't know what would happen
> if either the DR or BDR failed, forcing a re-election.  Hmmm, seems
> like a full mesh is safer!??
>
>     I prefer definately the full or partial mesh over the shared one.
>     shared
>     one mostly has a single link to the hub/switch and is a spof.
>
>     as for subnetting, choose a /30 netmask for each link. a /24 is a
>     great
>     waste of resoucres.
>
>
> Yes, exactly what I had in mind!
>
>     /30 handles exact 2 hosts (your router-link-adresses).
>     like 192.168.10.0/30 <http://192.168.10.0/30>
>     192.168.10.0 <http://192.168.10.0> network
>     192.168.10.1 <http://192.168.10.1> router 1
>     192.168.10.2 <http://192.168.10.2> router 2
>     192.168.10.3 <http://192.168.10.3> broadcast-adr
>
>     to your answer - make 1 subnet per link. in summary in your case: 5
>
>
> Thanks for the quick reply!!!
>
> -r
>
>
>     recvfrom@...
<mailto:recvfrom@...> wrote:
>
>     > Hi!
>     >
>     > While this question isn't strictly quagga-related, I hope I'm
>     not out
>     > of line
>     > asking this here...
>     >
>     > I'd like to modify my backbone area to create a partial mesh (all
>     > ethernet),
>     > as shown below:
>     >
>     > +--------+            +--------+
>     > |   R1   |-------------|   R2   |
>     > +--------+            +--------+
>     >      |        \               |
>     >      |          \             |
>     >      |            \           |
>     >      |              \         |
>     >      |                \       |
>     > +--------+            +--------+
>     > |   R4   |-------------|   R3   |
>     > +--------+            +--------+
>     >
>     > Right now R1-4 are in a single broadcast domain (but not with the
>     > topology shown above -- no loops, just a simple bus), using one
>     /24.
>     > With the above topology, I assume I need to further partition
>     the /24.
>     > My question... should there be four subnets or five (one per link)?
>     > Any great disadvantages to doing this, besides a slightly more
>     > complicated configuration?  I only see the upside.  TIA!!!
>     >
>     > -r
>     >
>     >------------------------------------------------------------------------
>     >
>     >_______________________________________________
>     >Quagga-users mailing list
>     >Quagga-users@... <mailto:Quagga-users@...>
>     >http://lists.quagga.net/mailman/listinfo/quagga-users
>     <http://lists.quagga.net/mailman/listinfo/quagga-users>
>     >
>     >
>
>     --
>     Christian Graf
>     Brombeerweg 2
>     37281 Wanfried
>     Mob. 0049.163.2474723
>
>
>
>
>------------------------------------------------------------------------
>
>_______________________________________________
>Quagga-users mailing list
>Quagga-users@...
>http://lists.quagga.net/mailman/listinfo/quagga-users
>  
>

--

-- 
Christian Graf
Brombeerweg 2
37281 Wanfried
Mob. 0049.163.2474723
Permalink | Reply | 0 comments |

Gmane