Multibeam Gpu Transient Pipeline for the Medicina Best-2 Array

Abstract

Radio transient discovery using next generation radio telescopes will pose several digital signal processing and data transfer challenges, requiring specialized high-performance backends. Several accelerator technologies are being considered as prototyping platforms, including Graphics Processing Units (GPUs). In this paper we present a real-time pipeline prototype capable of processing multiple beams concurrently, performing Radio Frequency Interference (RFI) rejection through thresholding, correcting for the delay in signal arrival times across the frequency band using brute-force dedispersion, event detection and clustering, and finally candidate filtering, with the capability of persisting data buffers containing interesting signals to disk. This setup was deployed at the BEST-2 SKA pathfinder in Medicina, Italy, where several benchmarks and test observations of astrophysical transients were conducted. These tests show that on the deployed hardware eight 20 MHz beams can be processed simultaneously for ∼640 Dispersion Measure (DM) values. Furthermore, the clustering and candidate filtering algorithms employed prove to be good candidates for online event detection techniques. The number of beams which can be processed increases proportionally to the number of servers deployed and number of GPUs, making it a viable architecture for current and future radio telescopes.

GPU-Powered Coherent Beamforming

“…Comparing this implementation with a similarly optimised CPU version of the same algorithm run on a dual Intel E5-2640 system shows an overall performance gain of an order of magnitude (∼10x). We have integrated this kernel with two real-time software pipelines at the BEST-2 array, a standalone beamforming pipeline and the transient detection pipeline developed by (Magro et al , 2013), and show that when the beamformer is used in conjunction with other processing kernels, and as long as the generated beams stay in GPU memory, beam generation overhead is minimal when compared to processes such as dedispersion. This can alleviate the PCIe data movement bottleneck for specialised pipelines.…”

“…Therefore a single 10 GigE link is sufficient for data transfer between the F-engine and GPU server. At the receiving end, the packet receiver used for (Magro et al , 2013) was updated to be compatible with this format. Groups of 128 time samples are considered as SPEAD heaps, and thus heap functionality is still applicable.…”

“…We have also integrated the GPU beamforming kernel into the GPU-based transient detection pipeline described in (Magro et al , 2013). This section provides general design considerations for creating a transient detection pipeline with beamforming capabilities, assuming that a single GPU is capable of running the transient detection processing stages for at least one synthesised beam.…”

“…The transient detection pipeline is then launched for each synthesised beam, keeping the beamformed data in GPU memory. The output of the transient detection pipeline is then processed on the CPU, as described in (Magro et al , 2013).…”

“…We also present a direct comparison with an optimsed Central Processing Unit (CPU)-version of the same algorithm. In Section 1.1 we present a beamforming system deployed at the BEST-2 array in Medicina, Italy, whilst in Section 3.2 we discuss the integration of this system with the the real-time transient detection system developed by (Magro et al , 2013).…”