[Click] Re: click Digest, Vol 29, Issue 17

[Marko Zec]

On Sunday 20 November 2005 21:04, Siddharth Kasat wrote:
> Hi,
>
> I wanted to measure the performance of IP lookup for Lulea trees and
> other kinds of look up.
> I see some measurement figures in
> http://pdos.csail.mit.edu/click/doc/IPRouteTable.n.html
>
> How do I go about it?
>
> Here is my idea about..I can send packets hard coded but then how do
> I make entries in routing table?


You may but do not need to generate valid packets in order to stress 
test the lookup elements.  Lookup elements take IP address annotation 
with each packet as its input, not the actual contents of the packet.  
Moreover, it might be a good idea to completely randomize the 
destination address annotation in performance tests.  Tests with 
localized traffic patterns might get serviced from the CPU cache with 
hit ratios unrealistically high compared to what could be expected in 
real life or worst-case scenarios (which is basically the same thing -> 
too oftern real life turns into the worst case scenario anyway :-)

Here's a basic configuration that I used in my performance tests:

rt :: RadixIPLookup(
216.60.228.0/22 64.200.95.239 1,
210.210.0.0/17 4.68.0.243 1,
... etc etc...
201.225.40.0/21 206.24.210.26 1,
);

RandomSource(4) -> GetIPAddress(0) -> rt;

rt[0] -> Discard;
rt[1] -> Discard;

Hope this helps,

Marko


> Any pointer in this regard would be a great help.
>
> Thanks,
> Siddharth
>