Tara Madhyastha - US grants

The storage density of rotating magnetic recording is approaching its
theoretical maximum. Magnetic probe-based technology avoids these limitations by using techniques such as orthogonal recording which promise very high density storage within the next five to ten years. Probe-based storage devices promise improved access times, enormous potential parallelism gains, and remarkable storage densities. However, because of the unique characteristics of these devices there is a high
probability that existing file system architectures and algorithms will be suboptimal. By reexamining these basic structures in the context of probe-based storage, it is likely that significant performance gains can be achieved.

The proposed work comprises fundamental research in four areas: simulation of probe-based storage devices, architectural issues such as parallelism and caching, storage allocation and file layout, and request scheduling. In reexamining these basic issues for this new technology, this research creates a body of work that will lead the way in the development of secondary storage systems for such devices. This research is likely to result in a better understanding of the implementation details associated with probe-based storage devices to provide a set of algorithms and structures that can be used in systems implementations employing them.

Despite the growing complexity of the storage landscape, tools to architect new storage systems are practically nonexistent. The purpose of this CAREER grant proposal is to develop a long-range program of research and education in design and performance analysis of storage systems, with a specific emphasis on emerging probe-based storage technology. The main question this research proposal asks is: "How do we design probe-based storage to give optimal performance for a workload with certain characteristics?" Answering this question requires a long-range research plan with thrusts in several areas, including creating physical models for the device, developing tools and methodologies for optimizing design based on a specific workload, and novel research in workload characterization to generalize our conclusions. Although our initial focus is on probe-based storage, we envision creating a parameterized, plug-and-play performance environment in which many other models may be used.

The education component of this proposal represents a philosophy of curriculum development to increase awareness of storage systems, and a general emphasis on fostering active learning skills.

The I/O demands of scientific applications are increasing exponentially as researchers strive for greater accuracy in models and simulations of physical systems, and as more information about the world is obtained. To deal with this aspect of computation implies that one cannot ignore the I/O gap and only concentrate on processor-centric issues. As a concrete example, the size of Genbank, a nucleotide database, has been doubling every 14 months and is currently 33 GB. In the commercial arena attention has turned to information and its location also. More and more network services, fueled by the demands of information-intensive Web applications, are shifting the focus in computer systems to I/O.

With the growing importance of data comes the need to address its management. The commodity cluster of the near future, which consists of a combination of network-attached storage, intelligent data appliances and workstations, and high-performance networks, is a highly parallel, heterogeneous computer. However, the file abstractions in use today cannot yet fully exploit either the resulting computational power or data locality.

This award examines how to improve I/O performance on such a platform for a class of important applications. UC Santa Cruz has a leading computational biology group with a 100+ node Linux cluster; the parallel sequencing codes that they use are an obvious target. Other groups have also expressed interest in this cluster; and its use will be an excellent source of applications.

There are three main research objectives. First, develop an I/O programming model that unites computation and storage. This will allow the programmer to express "storage operations" that can be readily parallelized. Second, build an adaptive infrastructure and develop analytic models to determine the optimal execution path, which may be parallel or sequential. Third, develop offline models to answer questions about how to allocate resources within this heterogeneous environment.

The goal in creating a new I/O interface is to associate computation and data so that code can be easily executed at the source of the data. For example, suppose we want to search for a word in a dictionary. The standard procedure would be to open the dictionary file, read it in chunks, search for instances of the word in each chunk, and close the file. This operation is sequential and processor-centric; the emphasis is on moving the data to the processor rather than a computation to the data. Instead, we propose to associate a "search for word" command with the dictionary in the form of a new I/O interface. Now the "search for word" becomes a high level abstraction for the same code described above, a single remote procedure call, or a parallel procedure (depending on the location of the dictionary).

The goal of this collaborative research project involving Christos Faloutsos and Ngai Hang Chang at Carnegie Mellon (award 0083148) and Tara Madhyastha at U of Cal Santa Cruz (award 0083130) is to develop and apply statistical and datamining tools to analyze bursty time sequences, with emphasis on I/O traffic optimization. The interdisciplinary team includes researchers in computer science, computer engineering and statistics, and industry collaborators. The approach has three parts: (1) advanced statistical tools using the ``ARFIMA'' method; (2) wavelets and the related ``80-20 law'' to model disk traffic; and (3) incorporation of these models
inside the so-called ``Active Disks'', with the goal to build self-tuning, adaptive disk subsystems.
The results will advance data mining and statistics as well as disk design. An easy-to-use toolkit "T-REX" will aid in I/O and systems design, handling bursty traffic, and better buffering and prefetching. The theory behind the T-REX toolkit will be based on new data mining algorithms and statistical methods that model self-similar time sequences (like web and network traffic,
in addition to I/O traffic). The research team has strong ties with database, data mining and disk manufacturing industrial groups, and this will aid in testing the research toolkit and its technology transfer. It can be expected that the T-REX system will significantly aid the design of disk sub-systems with beneficial impact on the storage industry.

Institution: University of California - Santa Cruz
Proposal Number: EIA - 0244016
PI: Richard Hughey
Title: REU Site in Information Technology at the Baskin School of Engineering

The Summer Undergraduate Research Fellowship in Information Technology (SURF-IT) program provides an intensive and personalized summer program for 12 underrepresented (women and minority) and disadvantaged undergraduates. The students are drawn primarily from institutions that do not have strong, personalized research opportunities. SURF-IT is a 10-week summer program that includes research directly supervised by a faculty member in computer engineering, computer science, or electrical engineering. Students also have weekly meetings on graduate school preparation, research ethics, and research presentation, as well as field trips to local neighboring Silicon Valley research laboratories. The summer program concludes with
a research poster presentation by the SURF-IT students, coordinated with other undergraduate summer programs within the institution.

The goal of the program is to promote diversity in engineering by assisting motivated students in obtaining graduate degrees to then become positive role models for future generations of college students throughout the nation. Thus after the summer program, the faculty maintains continuing contact with SURF-IT alumni to assist them in applying to and succeeding in graduate school, as well as in their eventual careers.

The objective of this award is to develop a computational framework for identifying the critical network topology of brain connectivity in neuroimaging data, specifically functional magnetic resonance imaging (fMRI). In this framework, network optimization modeling and mathematical programming algorithms will be employed to characterize connectivity patterns in fMRI data from different brain regions. Machine learning techniques will be employed to construct a pattern recognition model used to detect biomarkers and predict the brain disease conditions (i.e., abnormals vs. controls). An information-theoretic approach will be used to select the most informative brain regions to improve the generalizability and to increase the accuracy of the diagnosis prediction model.

If successful, the results of this research will lead to improvements in efficiency and efficacy of brain functional connectivity modeling and new developments of optimization methods for handling large-scale spatio-temporal data. The developed computational framework will be extremely useful for neuroscientists and neurologists to identify abnormal functional connectivity in the brain and to gain a greater understanding of the brain function. The framework will be employed and tested as a novel biomarker for differential diagnoses of brain disorders. Alzheimer?s disease (AD), autism spectrum disorder (ASD), and Parkinson?s disease (PD) will be the case points in this project to test if our computational framework is a sensitive enough tool to detect alterations in brain connectivity associated with brain disorders. Accurate diagnosis can substantially extend a patient?s lifespan and some treatments have different outcomes at different disease stages. Additionally, the developed computational framework can be applied to other real-life large-scale spatio-temporal data that arise in other research areas such as manufacturing, medicine, bioinformatics, neuroscience, finance, and geosciences.

Cognitive impairment is the most disabling non-motor feature of Parkinson Disease (PD) and causes the greatest degree of caregiver distress. The large majority of patients with PD will eventually suffer cognitive impairment, and although treatment of motor parkinsonism has improved, cognitive impairment has proven more difficult to treat. When cognitive impairment appears, it tends to have a profile of more affected and less affected domains that suggests differential regional cortical involvement. Measurement of cognitive performance through neuropsychological tests tells us what functions are impaired but not why. A better understanding of the causes of impairment would help identify therapeutic targets for cognitive symptoms. This would lay the groundwork for developing biomarkers of brain (patho)physiology that would advance the goal of precision medicine for treatment of PD. This project exploits the discovery that regional spontaneous cortical activity measured by fMRI at rest has a spatial structure that includes the same cognitive networks identified during tasks. These ?intrinsic networks? are altered in PD and many other diseases, providing an important physiological connection between brain structure and cognitive function. We have developed fMRI-based methods for sensitively quantifying differences in intrinsic networks and shown that networks differ in people with PD and controls. Although fMRI has good spatial resolution, because of hemodynamic delay it lacks temporal resolution. Therefore, differences in networks observed in PD may reflect differences in timing, or dynamics, that we cannot measure at the temporal resolution of fMRI. We want to integrate our innovative framework with a complementary modality, electroencephalography (EEG), which helps us to distinguish differences in spatial extent and timing of networks. Recently, we have shown a systematic relationship between intrinsic network activity and the time course of an attention network task that is different in PD and controls. This link between intrinsic networks and task-related activity allows us to ask how intrinsic network activity relates to cognition in PD. We hypothesize that alterations to networks that support attention and memory are specifically related to cognitive performance in these domains in PD, and that temporal information from EEG will help to resolve this. We will test this hypothesis by simultaneously acquiring fMRI data (for high spatial localization of network structure) and electroencephalography (EEG) data (for high temporal resolution) in subjects with PD and controls, as we pursue the following specific aims: (1) Analyze and compare intrinsic activity from PD and controls obtained at rest. (2) Analyze and compare intrinsic activity from PD and controls during a covert visuospatial attention task with and without a memory load. (3) Analyze trajectories of longitudinal change.