Abstract
Download Conference Presentation
The DNA-based data storage pipeline implementation requires massive data computation in pre- or post-processing steps. State-of-the-art solutions do not consider the system as a whole but focus only on the pure theoretical foundations of the DNA medium. For example, the design of dedicated error correction engines or specialized algorithms to counter the inherent noisy behavior of the DNA (i.e., the IDS channel) must devise proper cost yet energy-efficient computing solutions. Given the complexity of the DNA nature, we must rely on massive computing capability that is not currently available in on-premise standard workstations or low-end servers. To this extent, we might envision dedicated High-Performance Computing (HPC) facilities to tackle the many challenges of a thorough DNA-based storage pipeline creation. However, the TCO and the OPEX of an HPC facility largely depend on the hardware integrated within. We prospect to embrace the heterogenous acceleration computing paradigm to reach performance and energy benefits that currently are out of reach with off-the-shelf products. Specialized programmable and runtime reconfigurable engines like Field Programmable Gate Arrays (FPGAs), Graphics Processing Units (GPUs), and even Computational Storage Devices (CSDs) can target peculiar steps of the DNA-based storage pipeline. The analysis time can be reduced significantly. We demonstrate a rack-format proof-of-concept in a relevant HPC facility hostable in future DNA-storage-ready data centers.