I had to refer to the performance notes from a source code file, these are the results from one of my projects:

performance:

single threaded operation with multithreaded libs
@ 550 mhz:

400 million geoip lookups on 4 billion ip address
space takes 31 seconds, or 12,903,225 per second

Notice that the lookups were done over a wide range of ip addresses to make sure that cpu cache effects would play as little a part as possible in the algorithm.

Put it this way: the geoip overhead of processing a log file with 13 million hits in it would be about 1 second.

My db is at the country/region level.

If you are on SQL 2005 you could use the new covered index feature, creating a single index across all the columns being looked up.

That way the result set is fetched direct from the index not touching the database and it really improves lookup times.

Info here:
[odetocode.com...]

Unfortunately we are still on 2000 on one of the boxes, but looks like we will be upgrading from 100,000 records to about 2,200,000, this is why I am thinking of moving to the first part of IP logic and splitting into 255 or perhaps re-aranging the data in the way that top countries of choise(most visits) records are sitting on top and USING TOP Keyword to only get the first record and stop searching.

BTW, when using TOP Keyword does the DB Engine stops after it finds the first match or it still goes thru the rest of the rows?

Can you guys suggest a good place to get this DB?

I will appreciate your help.

Thanks,
Rajat