Re: [mpls-linux-general] Benchmarking mpls problems !?

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

On Mon, Mar 14, 2005 at 10:56:46AM +0100, Sasa Pavlek wrote:
> Did anyone tested MPLS network with netperf, netpipe, ping or some other=
=20
> network tool, becouse I had some problems with measuring MPLS throughput=
=20
> and round-trip-time.

mpls-linux is not optimized for throughput it is optimized for flexibility
and correctness, you would have seen that had you searched the mailing
list archives.  I would be happy to work with someone if they would
like to try and optimize mpls-linux for throughput.

> Network:
>=20
> Source           Ingress                                               =
=20
>             Ingress          Destination
>                 Egress                                                =20
>             Egress =20
>         =20
> ---------        ---------        ---------        ---------       =20
> ---------        ---------        ---------
> | ip   =20
> |--------|ip/mpls|--------|ip/mpls|--------|ip/mpls|--------|ip/mpls|----=
----|ip/mpls|-------=20
> | ip    |
> | node  |        | node 1|        | node 2|        | node 3|        |=20
> node 4|        | node 5|        | node  |
> ---------        ---------        ---------        ---------       =20
> ---------        ---------        ---------
>                    |                                  |               =20
>                |
>                    |                                  |               =20
>                |
>                    |                                  |               =20
>                |
>                 ---------                         ---------           =20
>             ---------
>                =20
> |ip/mpls|-------------------------|ip/mpls|-------------------------|ip/m=
pls|
>                 | node 6|                         | node 7|           =20
>             | node 8|
>                 ---------                         ---------           =20
>             ---------
>=20
> All nodes are connected with 100 Mbps links.
>=20
> I try to compare IP and MPLS throughput and RTT, but I got odd results.=
=20
> For IP throughput between Source and Destination with netperf, I got=20
> approximately 90 Mbps over 1,2,3,4 and 5 mpls node, but for MPLS=20
> throughput with same tool I got only 350 kbps.

What are the MTU's of your links?  Did you turn off debugging?  Did
you recompile with MPLS_DEBUG turned off?

> When I ping (ping -qfc 10000 -s 8 dest_ip) Destination from Source, for=
=20
> MPLS I got 100 times bigger min/avg/max RTT resutls  then for IP. But=20
> for MPLS is faster total ping time then for IP.
>=20
> One more problem: Why IP TTL is decrementing when transmitting thruogh=20
> MPLS network. I think that IP TTL must be unchanged in MPLS network.

Wrong. Read RFC3031. Is is a user-configurable option.  In mpls-linux
it is configurable via the 'propagate_ttl' NHLFE option.

>=20
> 		=09
> ---------	---------	---------	---------	---------	---------	---------
> | ip    | ------|ip/mpls|-------|ip/mpls|-------|ip/mpls|-------|ip/mpls|=
-------|ip/mpls|-------| ip	|
> | node 1|	| node 1|	| node 2|	| node 3|	| node 4|	| node 5|	| node 2|
> ---------	---------	---------	---------	---------	---------	---------
> 		    |				    |				    |
> 		    |				    |				    |
> 		    |				    |				    |
> 		---------			---------			---------
> 		|ip/mpls|-----------------------|ip/mpls|-----------------------|ip/mpl=
s|
> 		| node 6|			| node 7|			| node 8|
> 		---------			---------			---------

PS please make you ascii drawings fit an 80 col terminal.

--=20
James R. Leu
jl...@mi...