Accelerating All-to-All Protein Structures Comparison with TMalign Using a NoC Many-Cores Processor Architecture

Computational challenges for the one-to-many and many-to-many protein structure comparison (PSC) problem are a result of several factors: constantly expanding large-size structural proteomics databases, high computational complexity of pair wise protein comparison algorithms, and the multitude of pair wise comparison approaches used in the field. Advances in processor architectures, such as manyc-ore CPUs, have enabled them to support parallelism making them of interest in speeding up PSC techniques. We presentrckAlign, an implementation of the popularly used TM-alignPSC algorithm, designed for the Single-Chip Cloud Computer(SCC), an experimental processor created by Intel Labs. We developed a skeleton library, rckskel, and implemented amaster-slaves variant of TM-align to exploit the parallelism offered by the SCC. We evaluated rckAlign on the SCC and compared it with existing TM-align software running on a dualcore AMD CPU (2.4 GHz) and on a single-core Intel P54CPentium CPU (800 MHz). We observed an 11-fold speedup relatively to the former and a 44-fold speedup relatively to the latter. A key aspect of the performance of rckAlign on the SCC, is the almost linear speedup achieved with the number of SCC cores used as slaves. The method presented can easily be applied to other PSC algorithms and extended to running multiple PSC algorithms within the same SCC chip.