An introduction to scaling out packages using parallelism with the "Work pile" pattern, balanced data distributor and "Roll your own" techniques. This will also cover how to achieve elastic style scale out and how well different approaches scale across CPU cores, including scaling with hyper threading turned on and off and an investigation into what the the affects of L2 cache splitting. Comparisons will be drawn between achieving parallelism using SSIS and the query execution engine, specifically around parallelism start up costs, partitioning workloads between threads and the impact of serial regions on the throughput of parallel schemas.
Presented by Chris Adkin at SQLBits XI