[j-nsp] hardware lookup OT

[Hannes Gredler]

On Mon, May 19, 2003 at 01:50:39PM -0400, Richard A Steenbergen wrote:
| On Mon, May 19, 2003 at 07:30:54PM +0200, Piotr Marecki wrote:
| > Lightly OT , but i'm very curious . Given that IP2 is able to make 40
| > million L3 lookups per second  - so how many memory accesses it is doing
| > per lookup and how this alghoritm may looks like ? 
| 
| I believe the actual L3 lookup done with an mtrie (the same data structure
| as CEF) implemented in the IP2. I seem to recall someone saying that the
| Juniper implementation was 16-1-1-1-1-1-1-1-1 vs Cisco's 16-8-8, but they
| could have been full of it. If you're looking for something to run on a 
| general purpose processor, you should probably stick to the way CEF does 
| it.

it is neither a 16-8-8 nor a 16-1-1-1-1-1-1-1-1 mtrie;

the algorithm is documented in the patent filing:

http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&u=/netahtml/search-adv.htm&r=15&f=G&l=50&d=PTXT&p=1&S1=juniper.ASNM.&OS=AN/juniper&RS=AN/juniper

/hannes