[Click] Max size of IP tables?

[Egi, Norbert]

Hi,
 
I was just trying to find out how large IP tables looking up packets with random destination addresses (i.e. hoping to cause constant cache miss, thus accessing the main memory all the time) affects the forwarding performance. I've been using the RadixIPLookup element, but also managed to make the DirectIPLookup and the RangeIPLookup elements work in Linux kernel by setting vmalloc=512M. However, all of these elements work for me only upto 4K entries (all the /12 possible addresses), I manege to load scripts with 8K or even 16K entries in them, but when I try to forward packets I get the following message: "chatter: IPRouteTable: no route for 144.38.117.64", even though all the possible IPv4 addresses are covered. In addition, the forwarding rate in these latter cases is always a fixed fraction of the generated packet rate (that is, when I put 100Kp/s in the forwarding path I got ~12Kp/s out of it, while with 200Kp/s I got ~24Kp/s, and so forth upto a certain rate where the rate doesn't increase anymore). These phenomenons suggests to me that probably not the whole IP table is loaded and therefore some IP addresses can't be looked up, causing packets being dropped.
 
Has anyone successfully experimented with IP tables larger than 4K or is there a limit for the number of entries in these elements? Any idea why they don't load properly? (By the way, they seem to load properly, but upon packets being generated the "no route for..." message shows up.)
 
Regards,
Norbert