[Click] Re: Click trouble :)

[Eddie Kohler]

Hi Eirill,

> First:
> I have installed and compiled a router configuration for testing on a
> computer with linux 2.4.18. My routerconfig-file was generated by editing
> the make-ip-conf.pl script to fit my IP configuration. Both userlevel and
> kernel mode compiled without errors. When I run Click in userlevel,
> everything works well. But when I run Click in kernel mode, it seems like
> click only forwards icmp-packets (??). I can ping both the test machines I
> have connected to the Click router, but when I try to send UDP packets
> (netperf test), the host seems to be unreachable. The traceroute command
> doesn't work either (the packet arrives at the click pc but is not
> forwarded). The router configuration is almost the same in userlevel and
> kernel mode, I only changed the line about localhost in order to make it a
> userlevel configuration. Also, after some time, the ping command doesn't
> work either. The route to the destination is lost... If I uninstall the
> kernel mode, and install it again, ping does still not work. BUT, if I
> uninstall kernel mode, run click in userlevel, abort and run it in kernel
> mode again, ping works! (but not netperf, traceroute, .....) 

It sounds like you have ARP issues! The Click router generated by
make-ip-conf.pl handles ARP as well as IP routing. Did you plug in the
correct Ethernet addresses for your network cards?

> Second (and I don't know if you can help me on this... All hints and tips
> are welcome!!):
> In my project I am going to study performance on the click router. Is it
> possible to see the performance on each element in the router configuration,
> and how do I do this? What I would like, is a outprint of timestamps when
> the packet is sent to the input port of each element, or something like
> this.

Check out elements such as RoundTripCycleCount, SetCycleCount, and
CycleCountAccum, all of which are described on our Web site.  There are
other options too: TimestampAccum, ...

> I'm going to look into the filtering mechanisms in click, and would
> very much like to see if the Ipfilter element take up much processing time.
> Do you know of any documents/papers that look into the security mechanisms
> in click? I have read the papers on the click web site, but couldn't find
> any with special focus on this...

Not as such, no. IPFilter's running time depends heavily on its input
string. The decision tree it produces is pretty well optimized, but it
remains a decision tree exclusively consisting of equality comparisons --
no hashes, for example. The most extensive discussions of this are in my
thesis, and in the ASPLOS paper (linked off www.icir.org/kohler/pubs/ ),
which talks about click-fastclassifier's performance.

Thanks for your interest!
Eddie