Re: [mpls-linux-general] mpls-linux speed test.

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

These are cool results, I would very interested to see the results with
GigE interfaces.  I would be nice to see if these same results cam be
achived by others.  Also I think testing with traffic on _many_ LSPs
would be interesting.

At some point I would like to concetrate on LER speed as well :-/

Laters

On Fri, Feb 28, 2003 at 12:26:57PM +0000, Radu Dragos wrote:
> Hi all.
> 
> Read this if you are interestet in some mpls-linux forwarding speed  issues.
> 
> Based on a discution with Jim on the IRC channel: #mp...@ir...
> I run the following test using a DataQualityAnalizer(DQA):
> 
> The idea is to stress a MPLS enabled Linux box (AMD 850MHZ, 256M RAM, RH7.3, 
> linux-2.4.19) in order to verify the capability of the mpls-linux 
> implementation to forward large bandwidths of traffic.
> 
> This is the framework I used:
> 
> 			192.168.254.7		192.168.254.25	
> 			eth1 ------------------------>	analyzer interface (AI)
> 				^
> 		mpls-linux  |  fwd					 DQA
> 				|		 			   
> 			eth0 <------------------------	traffic generator interface (TGI)
> 			192.168.100.2		192.168.100.1	
> 
> All the interfaces are 100Mbps Ethernet.
> 
> 1. Small packets (IP only no MPLS labels)
>  TGI sends TCP/IP traffic src=192.168.100.1 dst=192.168.254.25
>      packet size 64 Bytes
>      transmitted data rate: 76Mbps
> At this rate, ~150000 packets/s the linux machine is flooded and CPU 
> overloaded  	
>     AI gets only ~174 packets/s
> 2. Large packets  (IP only no MPLS labels)
> TGI sends TCP/IP traffic src=192.168.100.1 dst=192.168.254.25
>      packet size 1500 Bytes
>      transmitted data rate: ~98Mbps
>  At this rate ~8200 packets/s the linux machine behaves well 
> AI gets all the packets at the same rate TGI is transmitting.
> 
> 3. Large packets MPLS enabled
>  TGI sends TCP/IP traffic src=192.168.100.1 dst=192.168.254.25(MPLS label 16)
>      packet size 1500 Bytes
>      transmitted data rate: ~98Mbps
> All the MPLS packets received at eth0 were forwarded to AI. No packet was lost 
> as you can see from /proc/net/mpls*  :
> 
> 0x40004000 718224/1064407968/0 gen 16 0 1 POP FWD(0x00000002)
> eth0    0       6
> eth1    0       28
> 0x00000002 718224/1061535072/0 2 PUSH(gen 16) SET(eth1,192.168.254.25)
> 
> No significant CPU load was observed.
> The mpls for linux implementation is able to forward packets at FastEthernet 
> speeds.
> 
> I poses Gbit Ethernet interfaces for PS. Unfortunately I only have one Gbit 
> interface on DQA. Therefore, for the moment I can not run tests at Gbit 
> speed.
> 
> It remains now to analise the results in order to establish the delay 
> introduced by the MPLS forwarding plane.
> 
> Also, it remains the issue of variable size packets. 
> 
> All the comments or suggestions are wellcome.
> 
> Radu
> 
> 
> -------------------------------------------------------
> This sf.net email is sponsored by:ThinkGeek
> Welcome to geek heaven.
> http://thinkgeek.com/sf
> _______________________________________________
> mpls-linux-general mailing list
> mpl...@li...
> https://lists.sourceforge.net/lists/listinfo/mpls-linux-general

-- 
James R. Leu