Scott, Ross and Greg talked about

Practical limits

Even a very efficient representation will eventually run out of room, but if we can get 4 biallelic snp per byte, we can represent Framingham SHARe? data (500k snps in 9600 subjects) in less than 2GB! Given that 64 GB will be increasingly common, the ceiling for any real data set is likely to be well within scope for a few years to come...

Practicalities

Need to make sure that all accessing of the efficient representation is itself efficient - no copying! - indexes rather than copies for subsets eg.

Resources and time

Moving Greg's subcontract from Rochester - possibility of increasing the level of support - Ross will check with the bean counters.