Accelerating Sequence Alignments Based on FM-Index Using the Intel KNL Processor

Abstract

FM-index is a compact data structure suitable for fast matches of short reads to large reference genomes. The matching algorithm using this index exhibits irregular memory access patterns that cause frequent cache misses, resulting in a memory bound problem. This paper analyzes different FM-index versions presented in the literature, focusing on those computing aspects related to the data access. As a result of the analysis, we propose a new organization of FM-index that minimizes the demand for memory bandwidth, allowing a great improvement of performance on processors with high-bandwidth memory, such as the second-generation Intel Xeon Phi (Knights Landing, or KNL), integrating ultra high-bandwidth stacked memory technology. As the roofline model shows, our implementation reaches 95% of the peak random access bandwidth limit when executed on the KNL and almost all the available bandwidth when executed on other Intel Xeon architectures with conventional DDR memory. In addition, the obtained throughput in KNL is much higher than the results reported for GPUs in the literature.

A survey on evaluating and optimizing performance of Intel Xeon Phi

“…However, KNL and KNM use out-of-order execution, high-bandwidth memory and interconnect. 10,53,73 Hence, they achieve high single core performance and overall throughput. Also, KNL/KNM support AVX-512 and all legacy ISAs such as x87/MMX/SSE/AVX/AVX2.…”

“…On KNC, attaining high performance may require significant code-rewriting such as programming in intrinsics, 7,18,20,40,52,53,59,82,84,86,87 although the need of intrinsics is not as stringent on KNL due to the improvements in microarchitecture, compiler and ISA compared to KNC. 83 Use of intrinsics is error-prone and increases code-development time.…”

“…Data layouts SoA, 20,22,23,28,30,36,50,79,82,90 AoS, 9 AoSoA 57,90 Data alignment 6,9,14,18,20,24,44,45,52,53,66,79,84,90 Padding 4,7,9,20,24,44,52,53,79,82,91 Dependency disambiguation 15,28,36,82,91 Prefetching Software, 4,7,9,14,17,22,23,40,41,50,…”

Genome Sequence Alignment - Design Space Exploration for Optimal Performance and Energy Architectures

“…Instead, simpler ARM based architectures along with high bandwidth memories can help in having an energy efficient architecture for NGS with better performance compared to existing solutions, leading to global energy savings for HPC data centers. NGS applications use full-text indexing strategies, such as the FM-index, based on Burrows-Wheeler transform (BWT) [18], [19] for fast sequence alignments. Additionally, Bowtie2 [20] is a new state-of-the-art NGS application based on FMindex, with efficient multi-threading capabilities.…”

“…Then, HISAT2 [22] is a graph based sequence alignment application and superior to both Bowtie2 and BWA-MEM in performance. HPC class compute resources like the Intel Xeon Phi KNL processors [23], Intel Skylake and Broadwell architectures, which support multiple threads, are used to run these NGS applications and maximize their performance [18], [24]. GPUs are also explored for genome sequencing as in [25].…”