[Click] click classifier

[Eddie Kohler]

Hi Tomas,

Tomas Mahoney wrote:
> hello clicksters...
> 
> i need to use about 3000 say basic patterns in the
> IPClassifier (Classifier) element. 

Wow!

> i will probably run some tests but in the meanwhile, 
> 
> - any ideas about much this will affect overhead or
> slow down click?

It will be expensive -- particularly if a lot of the patterns see a lot of use.

> - do u think the classifier element will be able to
> handle several 1000 patterns or was it not built for
> that? any ideas on potential ways to improve it for my
> purpose?

It will handle them, but with poor algorithmic complexity (worst-case linear 
in the number & length of patterns).

> - how does changing the size of MAX_DLIST, the bounded
> number of paths per state, affect performance?

Probably not enough.  Depends on the patterns; best way is to give it a try!

> - what is the relation between IPFilter element and
> Classifier?

IPFilter is just a different syntax veneer for Classifier.  IPFilter and 
Classifier share all the same optimizations, for example.  IPFilter's raw 
push() method is a tiny bit different, in that it loads the IP header pointer.

> - it appears the classifier element uses some
> dominator optimization stuff which i have no clue
> about and am just starting to read up on. any tips on
> how the dominator algorithm stuff is applied to
> classifier?

Oh, dear.  Wish we had a better writeup of this stuff; it is not uncool.  But 
really if you have 3000 firewlal-type patterns you want a completely different 
algorithm, most liekly hashtableish.

Eddie


> 
> thanks 
> tomas
> 
> 
> 
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around 
> http://mail.yahoo.com 
> _______________________________________________
> click mailing list
> click at amsterdam.lcs.mit.edu
> https://amsterdam.lcs.mit.edu/mailman/listinfo/click