You could try adjusting the values for DefaultBUfferMaxRows, DefaultBufferSize and EngineThreads, but it's something of a black art....I've found, anyway. Also, minimise the number of 'blocking' tasks in your package (sorts for example) that have to wait for the entire data set to flow through them before any rows can be passed to downstream components.

If you have a way of dividing the total data set into equal chunks you could run mutiple Data Flow Tasks in parallel. Oh and look at increasing the packet size too.

Have a look at this TechNet article

Regards

Lempster