SoCs Validation and Evaluation
Task 1.3.Validation of the derived architectures using large scale biomolecular networks
All designed parallel architectures have been described in fully parametric HDL code that can be used to directly synthesize a multiprocessor SoC implementation with desirable characteristics for reconfigurable hardware (FPGA). In this way we can synthesize and validate SoC realizations of different complexity, matching the underlying characteristics of the biomodel(s) they are called to stochastically simulate (i.e. number of species and reactions, max. reactions order etc.). We used the flexibility provided by the soft IP cores developed in tasks T1.1 and T1.2 to create and validate a library of different SoC instances and use each time the appropriate instance to simulate a large variety of biomodels of increasing complexity available in the literature (in standard SBML format). We used benchmark (synthetic) biomodels, with 2nd and 3rd order reactions, in which the number of reactions and the dependencies among them can be prescribed in a systematic manner. Such models are needed to perform performance scalability studies. In addition we have simulated several published biomodels developed by our group or others, including models of multi-cellular and reactive-diffusive biological systems that can also be scaled up to give very large networks as the cells population increases. In a medium size modern FPGA (Xilinx Kindex 7) we could fit up to 8 PEs for the NRM-SSA and 16 PEs for the FRM-SSA, each PE capable of handling up to 2K and 4K reactions respectively. There are really no scalability restrictions imposed by the SoCs architecture since the inter PE communication during simulation is minimal by design and the bottleneck is only the communication with the host PC, which can be controlled by lowering the sampling rate of simulation results (a user defined parameter).
Deliverable (technical report)
D1.3 Validation of the derived FPGA implementations using biomodels of increasing complexity