[Bridge] Benchmarking bridging vs. routing on same hardware/network

[Stephen Hemminger]

On Wed, 26 Jan 2005 14:52:13 -0600
<dan-linuxbridge at unpossible.com> wrote:

> I have been running benchmarks to compare the performance of bridging to
> routing on the same hardware.  
> 
> The bridge/router system is a dual Opteron 246 (2.0Ghz); Tyan 2882
> motherboard; 2 GB PC3200 memory (1 GB in each CPU's bank; 128bit
> configuration); dual e1000 based NIC; PCI-X 133Mhz; 2.6.9 Kernel; NAPI
> enabled in the kernel and the NIC drivers; CPU affinity ties each eth to a
> specific CPU; system otherwise idle.  
> 
> My test configuration is:
> 
> Packet Generator System <--> Bridge/Router <--> Packet Recipient
> 
> The network and hosts are otherwise idle.
> 
> Using pktgen (the Linux packet generator) I can send 459,000pps through this
> hardware when it is configured as a router.  I only get 252,000pps through
> the hardware when configured as a bridge.  
> 
> Needless to say, I use two IP subnets when routing, and a single subnet
> while bridging.  Also, when routing I set the mac addr destination for
> pktgen to the router interface (otherwise it won't pick up the packets),
> whereas when bridging I set it to the recipient's mac address.
> 
> Any ideas why the performance of the bridge is so bad?  

No, but perhaps oprofile would help.


-- 
Stephen Hemminger	<shemminger at osdl.org>