The Science Division is developing scale-up and hybrid analytics technologies to help medical professionals extract knowledge and insights from large, complex genomic databases to improve Precision Medicine for a number of diseases. Scale-out supercomputing often involves linking together lower-performance machines to collectively do the work of a single scale-up machine. Scale-out parallelization has not proven to be cost-effective or provide the high performance needed to analyze very Big Data in real time, due in large part to the latency of processing massive datasets at network speed.
In scale-up supercomputing, massive datasets are analyzed completely in RAM at compute rather than network speed–minimizing latency issues–resulting in orders of magnitude speed and performance improvements. The limitations of clusters to perform analytics on very large data sets in-memory and scale without severe latency implications caused by the cluster networking have led to a revived interest in scale-up platforms. Hybrid supercomputing utilizes the unique advantages found in both scale-out and scale-up architectures, while also employing different processor types (e.g., CPU, GPU, FPGA) to accelerate speed and performance of different steps in a process.