First Annual CROSS Research Symposium
First Annual CROSS Research Symposium
& 8th UCSC Systems Oktoberfest
Monday/Tuesday, October 24-25, 2016
Baskin School of Engineering, University of California, Santa Cruz
The Center for Research in Open Source Software (CROSS) aims to integrate the activities of open-source software communities into university research and education. The Center uses open-source projects as a way to share research and talent, creating an excellent learning and research environment.
The CROSS two-day Symposium provides an opportunity to learn about CROSS projects, interact with CROSS faculty, graduate students, and affiliated researchers, as well as outside researchers and members of the OSS community. In keeping with tradition, the UCSC Systems Research Lab will also host its annual Oktoberfest barbecue dinner on the evening of the first day of the symposium.
This year, the symposium will include four workshops on the following topics: (1) Secure, Real-Time Sharing of Cancer Gene Information, (2) Workshop on Data-Driven Dynamic Networked Systems, (3) Programmability and Architecture of Storage Systems, and (4) Towards a Better Computer Arithmetic: A unum (Universal Number) Workshop.
Each day begins with a plenary session and continues with two tracks of topic-specific workshops.
Cancer is a disease of the genome. Thanks to new genome sequencing technology, we have seen breath-taking advances over the last ten years in our understanding of which genomic mutations accumulate in cancers. However, we also realized how big the sheer number of possible mutations is in the three billion letters of our genome. For the majority of the mutations we do not know if they drive uncontrolled cell growth or are rather mere passengers that have little to no effect. To understand these mutations we need data, but today researchers are limited to small curated datasets in the thousands from participants from the past. We need millions of datapoints as a reference and new datapoints in weeks for prospective participants. In this workshop we'll cover the basics of cancer genomics as well as emerging approaches to real-time sharing of genomic data globally, bringing the compute to the data to retain privacy, optimizing large scale genomic computing for speed and managing both the data and compute.
The performance and robustness of a vast majority of systems of the future, such as autonomous aerial systems, self-driving cars, internet of things, and cloud computing, will rely on algorithms that efficiently process large amounts of data. Such is the case when the algorithms are in charge of making decisions according to (dynamically changing) measurements obtained from networked sensors, mobile agents, and users. In these systems, the storage and computing requirements can be quite expensive and a limiting factor in their deployment. The goal of this workshop is to showcase advances obtained by researchers from industry and academia that tackle key challenges in the development of such dynamic networked systems emerging in autonomous systems, internet of things, smart storage and cloud computing research.
Tuesday, Track 1:
Programmability and Architecture of Storage Systems
(chairs: Carlos Maltzahn, Jishen Zhao)
In 1936 Alan Turing laid the groundwork of the stored-program concept, an important milestone in the evolution of digital computers. 80 years later we still treat an important subsystem of digital computers essentially as a fixed-program computer: storage systems. Since fixed-program storage systems cannot be easily reconfigured, application programmers have to frequently take heroic measures to achieve acceptable performance, trying to align their storage access strategies to "magic numbers" that are often idiosyncratic to any particular storage system. There is growing consensus that this approach is no longer feasible due to the increasing heterogeneity of storage device architectures and the - often conflicting - requirements for high throughput, energy efficiency, and resiliency, along with long lifetimes. Future memory and storage systems can employ a combination of cost-effective devices, such as flash, disk, and emerging NVRAM technologies, but it takes often years to bring new storage systems up to production quality. This workshop is intended to discuss "programmable" storage systems. From the application programmer's point of view, programmability means an API to express access needs without having to know about the internals of a storage system. From a storage systems designer's point of view, it means an API to leverage a diversity of devices to achieve high-performance, energy efficiency, and lifetime enhancements without the need to create a new storage system from scratch.
A big issue facing cutting-edge computing is power consumption. An age-old issue is the problem that inexact arithmetic introduces into algorithms.
A new arithmetic formulation, involving unums (universal numbers) and SORNs (Sets of Real Numbers), has been proposed as an alternative to traditional floating point. A unum not only contains information about its own accuracy, but also then has the potential to reduce the number of bits required to achieve a required accuracy, thereby addressing each of the above issues. The unum approach is in its infancy; it is both a new format and a new computing environment, and methods of problem solving require reconsideration when using unums. The workshop here then is aimed at educating interested parties on the concept, discussing the pros and cons, elucidating applications, and considering the way foward for practical implementation.
To register, go to REGISTRATION
Presentation Slides Upload Form - For Presenters Only