Re: [mpls-linux-general] really - really need help

[Igor V. <ch...@fl...>]

Hi there,

We're doing almost exactly the same thing. We've tested it on
LER1-LSR-LER2 configuration, kernel 2.4.17, mpls 1.128 (which seems to be
working much faster than mpls 0.996).

We have overloaded the LSR with some additional traffic, and measured the
packet loss, burst UDP traffic from LER1 to LER2. We have found that MPLS
packets are routed faster than IP packets, but only for some packet
load sizes (about 40-80 bytes and 500-1440bytes). For all other cases IP
was faster.

The maximum speed at which the burst traffic is sent depends on the
packet size as well (about 10Mbit/s for 1 byte load size, approaching
100Mbit/s for 1000+ bytes load size); MPLS packets are sent somewhat
slower than normal IP pakcets, but slower for *all* cases (so we guess
that doesn't really explain the confusing results at MPLS vs. IP packet
loss).

We're still waiting for James' opinion on this.

In the meantime, we have tried the new LDP patch for mpls 1.128. We have
managed to patch and compile it but the cvs version of ldp-portable seems
to contain typos! In ldp-linux.c file there are several occurances of
"ntohul" function calls. We couldn't find such a function anywhere, so we
changed it to "ntohl" and then it compiled properly. (!)

But, still something's wrong with it. We've tried to test it on a simple
LSR-LSR configuration but the session deletes soon after it's established:

prompt>add global 192.168.19.205
Adding interface lo
Adding interface eth0
Adding route 192.168.19.204/32 via 2
Adding route 192.168.19.0/24 via 2
Adding route 0.0.0.0/0 via 2
Adding route 192.168.19.205/32 via 1
Adding global object with LSRID c0a813cd
prompt>add interface eth0
Adding interface eth0
prompt>TCP connect
Adding in label: 16 
Adding in label: 17 
Adding out label: 17 key: 0x5 if_index: 2
Binding FEC: 192.168.19.204/32 to label: 17 key: 0x5
Deleting in label: 16
Deleting in label: 17
Unbinding FEC: 192.168.19.204/32 from label: 17 key: 0x5
Deleting out label: 17 if_index: 2
session delete

...and the strangest thing, the FEC which is displayed in the deamon isn't
the one used :

voip5:/home/igor# cat /proc/net/mpls_out 
0x00000002 7/490/0 1 PUSH(gen 17) SET(eth0,1.0.0.0) 
0x00000003 8/560/0 1 PUSH(gen 17) SET(eth0,180.0.7.8) 

So we're additionaly puzzled. Fortunately, the ldp-portable we grabbed
with cvs in April works fine with 2.4.12 and mpls-linux-1.0.

Hopefully someone can shed some light on this? 
Thanks in advance


Igor Vukomanovic
Zeljan Sudeta




> Hi all,
> We are working on mpls for our final project and we have installed mpls-linux-1.0 in 4 PCs successfully using red hat 7.2, kernel 2.4.7, and 3com NIC.  As we knew that mpls work better than ip if the network busy. But until now our ip work better than mpls.
> Our topology :
> 
> Host A - LER A --- LSR A --- LSR B ---- LER A --- Host B
> 
> First we use clean network, and we ping host B from host A, and the result IP better than MPLS.
> Second, we generate trafik (we use mtools that made by Stefano Avallone - thanx Stefano) in LER A to LSR A in order to make the network a little bit crowded, than we ping host B from host A and still IP better than MPLS.
> Third we generate trafik from LER A to LSR A and LER B to LSR B to make the network more crowded, than we ping host B from host A still IP better than MPLS.
> 
> So, can anyone told us how to show that MPLS is better than IP ? what make mpls-linux different than IP (or better than IP) ?
> Because our purpose to work on it is to prove that MPLS is better than IP
> 
> Thank you very much, and we waiting for any suggestion and advice
> 
> Venda, amiek, andy