Re: [mpls-linux-general] restricted amount of labels

[James R. L. <jl...@mi...>]

Your using the most recent version from CVS right?

On Fri, Dec 07, 2001 at 11:17:49AM +0100, Kauder Thorsten wrote:
> Hi Everyone,
> I managed to get the LDP (linux-port) working and found out that one 
> has to be very carefull using it.
> A single wrong input sometimes causes a segmentation-fault.
> 
> Here's the scenario I established:
> 
>      dummy0 --------- eth0   eth0 ------- eth1   eth0 --------- dummy0
>       <-----| LER 1 |-------------| LSR  |------------| LER 2 |---->   
> 20.0.0.0/24 --------- 10.0.0.0/24 ------- 11.0.0.0/24 --------- 20.0.0.0/24
> 
> I have zebra running on all machines and propagate virtual routes pointing out the dummy devices using OSPF.
> After the Routing-tables are correct on all mashines by typing:
> 
> 	./ldp_linux
> 
> after hitting enter twice the prompt appears.
> 
> On LSR:
> 
> I add a global object at the LSR with the LSR-ID 10
> 
> 	add global 10

LSRID should be an ip address.  On a Cisco the LSRID is the highest
IP address attached to the router, or an address specified by the
user (or the address of the first loopback interface).  So make the
LSRID one of the addresses of the linux box.  You should also make sure that
the LSRID is advertised as a /32 to all other routers.

> then enable LDP on both interfaces
> 
> 	add interface eth0
> 	add interface eth1
> 
> On LER1:
> 
> 	add global 2
>  	add interface eth0
>  
> On LER2:
> 
> 	add global 3
> 	add interface eth1
> 
> I found out that this order is the safest.
> What happens now is that they really exchange labels wich is fine!! but I still have some problems:
> - If I change the routing table while LDP is running the label-mapping stays unchanged. ( if I got you right. with ldp_zebra this should work )

In theory.

> - you're adding interfaces by their names but delete them by ID. deleting with the name as argument causes a crash quite often.

You can only delete things by their index.  ldp-portable keeps track of
things by indexes, not names.

> - according to /proc/net/mpls-in a forward on LSR is only assigned to the routes pointing to the LER who was activated first ( here LER1 ). 
> 	 in other words the mappings aren't updated if a new LDP-Router is added.
> That's all not so important for my purposes. It works somehow.....

Hmmm.  I plan to spend sometime working on ldp-portable once I finish up
my most recent set of change to mpls-linux.  Maybe at that time I can try
to duplicate your setup and help debug it.

> The thing I could really use is support for lots of label-mappings. Right now, after a bit more than a hundred label-mappings there 
> appears one partial labelmapping (only FEC and label no op-code..) Then nothing.
> This occurs not only when using LDP but also when mapping the labels using the mplsadm tool. So I guess this is rather a mpls 
> than a ldp problem. I really need at least 1000 labels for test-purposes.
> Would that be possible or is there a lack of memory somewhere ???

Hmmm.  very interesting.  I haven't done any mpls-linux stress testing yet.
I will do some testing for this.

> greetings
> Thorsten

Thanks you for your feedback.  I'll see what I can do to help.

Jim
-- 
James R. Leu
jl...@mi...