S-Aligner: Ultrascalable Read Mapping on Sunway Taihu Light

The availability and amount of sequenced genomes have been rapidly growing in recent years because of the adoption of next-generation sequencing (NGS) technologies that enable high-throughput short-read generation at highly competitive cost. Since this trend is expected to continue in the foreseeable future, the design and implementation of efficient and scalable NGS bioinformatics algorithms are important to research and industrial applications. In this paper, we introduce S-Aligner–a highly scalable read mapper designed for the Sunway Taihu Light supercomputer and its fourth-generationShenWei many-core architecture (SW26010). S-Aligner employs a combination of optimization techniques to overcome both the memory-bound and the compute-bound bottlenecks in the read mapping algorithm. In order to make full use of the compute power of Sunway Taihu Light, our design employs three levels of parallelism: (1) internode parallelism using MPI based on a task-grid pattern, (2) intranode parallelism using multithreading and asynchronous data transfer to fully utilize all 260 cores of the SW26010 many-core processor, and (3) vectorization to exploit the available 256-bit SIMD vector registers. Moreover, we have employed asynchronous access patterns and data-sharing strategies during file I/O to overcome bandwidth limitations of the network file system. Our performance evaluation demonstrates that S-Aligner scales almost linearly with approximately 95% efficiency for up to 13,312 nodes (concurrently harnessing more than 3 millioncompute cores). Furthermore, our implementation on a single node outperforms the established RazerS3 mapper running on a platform with eight Intel Xeon E7-8860v3 CPUs while achieving highly competitive alignment accuracy.