Genomic Analysis at Scale - Mapping Irregular Computations to Advanced Architectures

Explore genomic data analysis at scale in this comprehensive lecture from the Society for Industrial and Applied Mathematics. Delve into the challenges and opportunities of mapping genomic analysis problems to petascale and exascale architectures. Learn about high-performance tools for analyzing microbial data, including alignment, profiling, clustering, and assembly. Discover common computational patterns and motifs that inform parallelization strategies and architectural requirements. Examine two general approaches to genomic analysis: asynchronous one-sided communication in UPC++ and bulk-synchronous collectives. Gain insights into specialized algorithms, GPU optimization, distributed algorithms, and communication-avoiding techniques. Understand the impact of growing genomic datasets on memory and computational requirements, and explore solutions for large-scale parallel platforms.

Introduction
Science Questions
Large Machines
Exoscale Computing
Specialization
Deep Learning Algorithms
Cost of Data Movement
Decoupling between GPU and CPU
Parallel algorithms
Metagenome assembly
Local assembly
Parallelization
Applications
Architectures
Hashing
Distributed Algorithm
Multicore Camera Counting
Minimizers
Alignment
GPU Optimized
Metahammer
Generalized Nbody
Long Read Overlap
Long Read Alignment
Communication Avoiding Algorithms
Bulk Synchronous Algorithms
Takeaways
QA