Collaborative Research: PPoSS: LARGE: Panorama: Integrated Rack-Scale Acceleration for Computational Pangenomics

点击翻译按钮获取中文摘要

Semiconductor technology scaling is slowing, and as a consequence, computer systems must increasingly rely on a heterogeneous mix of general-purpose and specialized computing engines. At the same time, computer users are tackling new problems with massive datasets that must be manipulated in irregular and rapidly changing ways while maintaining strict privacy guarantees. Efficiently supporting big, sparse, dynamic, and private data to solve large complex problems on heterogeneous systems is one of the grand challenges in software/hardware systems research. To address this grand challenge, the Panorama project is exploring integrated rack-scale acceleration for computational pangenomics. Integrated rack-scale acceleration refers to an emerging computer-systems paradigm that uses tens of tightly integrated computing nodes, each of which includes a mix of general-purpose processors and specialized accelerators interconnected with a special-purpose network. Computational pangenomics refers to a recent trend towards representing genomes, the genetic material of an organism, not as a single linear sequence of DNA base pairs but instead as an intricate network of sequences that efficiently represents the relationships between many individuals' genomes at once. Computational pangenomics naturally captures the trend towards big, sparse, dynamic, and private data and is thus a perfect application domain to explore heterogeneous software/hardware systems research. The project's novelties are: a truly cross-stack approach spanning applications, programming languages, compilers, architecture, security, and privacy including use of a one-of-a-kind Panorama prototype system; new hardware techniques to accelerate domain-specific computing and to unify heterogeneous systems; new software techniques to let programmers harness the performance advantages of heterogeneous systems; and new software/hardware techniques to make such heterogeneous systems more secure. The project's impacts are: to specifically enable computational biologists to better see the "genetic dark matter" of population-wide genomics which has been to date hidden, opening up new scientific discoveries; and to more generally enable future computer users to more easily take advantage of heterogeneous computer systems to solve large and complex problems. This project is also pursuing two broader impact initiatives. The first is an ambitious yet concrete initiative to increase participation of under-represented minority students in computer science by developing a low-level computer-systems module for a new four-week summer program targeting rising sophomores. The second involves specific plans to grow the open-source software/hardware ecosystem in the computational-biology and computer-systems communities.The Panorama project includes a highly interdisciplinary team of researchers across four focus areas: applications (computational biology), programming languages & compilers, computer architecture, and security & privacy. The team is taking a holistic software/hardware co-design approach to explore five tightly interconnected research thrusts. The first three thrusts are structured from top-down across the computing stack. Thrust 1 investigates new computational pangenomics data structures and algorithms and will develop PanoBench, a new benchmark suite suitable for driving the remaining thrusts. Thrust 2 investigates new programming-language and compiler techniques. Thrust 3 investigates new computer architectures with support for a whole-rack manycore with 1M+ cores and a partitioned global address space, unified array-based accelerators, and application-specific accelerator chiplets for computational pangenomics. The final two thrusts cut across both software and hardware. Thrust 4 investigates new security and privacy techniques including scalable secure computation on heterogeneous rack-scale systems, secure rack-scale resource management with auto-tuning, and differential privacy and homomorphic encryption for pangenomics. Thrust 5 involves holistically evaluating the research ideas in the other thrusts through the use of a one-of-a-kind Panorama prototype system.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.