[sounddust]

I simply don't believe that this model scales "over giant amounts of data", at least in any sort of real-world usage scenario. Can you back up this claim with benchmarks?