I think there has been a dangerous tendency to make things human readable, and I would argue that storage and retrieval should be optimized for the computer that is doing the job. Generating a report for humans would be useless with a million rows, and I have found that that the processing cost in rendering reports for human consumption has not been a bottleneck, the key is in applying general logic and business rules with explicit test cases, rather than the more common approach of applying eyeballs to long lists to figure out what was missed. since the libraries to manipulate IP addresses in Binary are readily available, I would stay away from char based approaches. I would also avoid the conversion to tinyint, because this is a transformation of the bytes and therefore should require and extra processing step for logical manipulation. If text based logs are the source for such data, why not turn them off, and write a direct Binary file for orders of magnitude faster ETL?

and following Mr Poole, With Intel's new 80 core processors will we have OCTDECAint?

Cheers,

John R. Hanson