Tim,

> > due to prefix compression within the index structures
> Ah, yes, I did write one of those once - a 1980s spelling dictionary for a word
> processor, where being able to encode an entire word in the smallest
> possible number of five-bit codes was *much* more important than being
> able to read the list backwards. (I think we ended up with less than 3 bytes
> per word for an English dictionary?)
>
> I'm slightly surprised that it's thought to matter with this century's disk prices
> however.

There are more "costs" that just disk space usage which need to be considered:
- Data Transfer speed from disk to RAM
- Data transfer from RAM to CPU
- RAM consumption of cache
- Loss of cache efficiency due to higher number of pages used by index structures

I do agree though, that if everyone was using SSDs for storage, the weight/strength of the above would be much weaker.


Sean