Abstracts

Storage Management - Where do CIO Really Spend the Money in Storage? (Abstract) Ealan Henis, IBM Haifa Labs We notice three trends: personnel management costs have become dominant, the blade approach to servers is gaining momentum, and networked technologies proliferate. To lower management and administrative personnel costs, the increasingly complex networked system requires simple and integrated management. We present HRL's vision and recent contribution in this area and its implementation within IBM system/storage management products. Object Storage - The Future Building Block for Large Scale Storage Systems (Abstract) Dalit Naor, IBM Haifa Labs A key development in the evolution of storage systems in the past two decades has been Networked Storage which migrated the data away from the computing entity. Application clients access their data via a file server that performs the access to the data over the network on behalf of the client. Today's Storage Area Networks place the storage *directly* on the network and allow, in principle, direct and non-mediated access to the data by the application client (thus removing the centralized file server from the critical I/O path). However, this promise can not be fully realized since security, data sharing management and other performance optimization must still be handled at the centralized server. Object Storage, a new emerging storage building block, was designed to address these difficulties by moving some of the file server layer funcitonalities to the storage layer. An Object Store raises the level of abstraction presented by a storage device. Instead of exposing a flat array of blocks, it maintains a collection of objects. The object store provides ``fine-grain,'' object-level security, improved scalability by localizing space allocation, and improved management by allowing end-to-end management of semantically meaningful entities. By doing that, it removes the file server data access bottleneck, but requires the storage device to perform sophisticated functions such as access control. In this talk I will describe what Object Storage is and what is it good for, and review IBM Haifa Lab's activities in this area, including building object stores and developing a standard. Search and Storage - Coping with One Billion File Filesystems (Abstract) Benny Mandler; Michael Factor; Miki Herscovici, IBM Haifa Labs As file systems grow in capacity and in the amount of files they can store, the scalability problem is shifted from that of storing files to that of finding stored files when they are needed. In this talk, I will call for augmenting the currently used physical rather rigid directory structure with a semantic view of the information stored in the file system. The semantic view will support context-sensitive search as well as aided browsing such that users can not only get help in locating a particular file that they are interested in, but also get help in organizing all available information from the point of view of interest at a specific moment in time. Thus, they will be able to find all the relevant information concerning a particular topic of interest, that is stored in the file system. A demo of an existing prototype will complement the talk. Disaster Continuity for an eBusiness World (Abstract) Aviad Zlotnick, IBM Haifa Labs I will go over the definition of business continuity and its importance. I will explain the different types and levels of business continuity, and the various solutions there are to make sure that the business continues to operate even after its data was hit. I will talk about Metro Mirror and Global Mirror–the two solutions developed mainly here in Haifa for solving the very difficult problem of business continuity. Building Scalable and Highly Available Systems - A Layered Approach (Abstract) Eliezer Dekel, IBM Haifa Labs The importance of guaranteed end-to-end high availability (HA) increases as business-to-business (B2B) interactions become progressively more sophisticated, lengthy, and involving more sites. Business' requirements for service resilience, and specifically, availability, are increasing. This heightened demand for HA information systems is contrasted to the fact that these solutions are currently created by the heroic efforts of a small group of high availability experts. The relative immaturity of the IT industry in building highly reliable systems, coupled with the rapidly increasing complexity of IT solutions, requires the adoption of more formal design analysis techniques and tools. In this talk, I describe a layered approach for building performance oriented scalable HA systems. Machine Learning - Theory and Practice (Abstract) Shai Fine, IBM Haifa Labs Machine Learning is an emerging field of research, which is attracting growing interest both within IBM and worldwide. In this talk I will review the principal notions of machine learning, and demonstrate how they can be put to practical use, by visiting some of the projects conducted by the Machine Learning team in HRL. Probably the most unique aspect of our work relates to the development of learning algorithms and their application for simulation-based hardware verification. Other activities we are engaged in, which I will briefly mention, include (but are not limited to) automatic enhancement of ranking provided by search engines, the prediction of query difficulty, learning aspects of selected Bioinformatics issues, such as Peptide identification based on learnt fragmentation rules, and anomaly detection and fault prediction in computerized systems. Introduction to Formal Verification and Its Applications (Abstract) Cindy Eisner, IBM Haifa Labs Formal Verification is a generic name for a variety of methods used to prove the correctness of systems or programs. Given a system and a formal specification of its desired behavior, formal methods can be used to mathematically prove that the system adheres to its specification, or to produce a counter-example showing the contrary. In this talk, I will provide a brief introduction to one type of formal verification, called symbolic model checking, and show how it is used to formally verify hardware and software. I will also discuss some unconventional applications of the technique, and discuss its limitations. Compiler Optimizations to Get the Most out of your SW/HW (Abstract) Ayal Zaks, IBM Haifa Labs A processor is only as good as the software it runs and compilers are responsible for translating your software efficiently to the processor. In this talk, I will review several compiler optimizations contributed by HRL to GCC (GNU Compiler Collection) from the past and present, including the inter-block instruction scheduler, software pipeliner, and auto-vectorizer. The GCC project of the Free Software Foundation is one of the most widespread and multi/re-targetable compilers in use today. Since 1987, many volunteers from academia, industry, and the private sector have been working to continuously improve the functionality and quality of GCC. Some of the compiler's key components were, and continue to be, developed at the IBM Research Labs. These contributions help improve the overall performance of code generated by GCC, and in particular, enhance IBM's RISC architecture. On-Chip Interconnect-Aware Design and Modeling Methodology, Based on High Bandwidth Transmission Line Devices (Abstract) David Goren, IBM Haifa Labs As the operation speed of integrated circuits reaches the multi-gigahertz range, the impact of on-chip interconnect on design performance has become a major issue. This subject has today become a major theme in design automation conferences, such as DAC and DATE, and is considered to be a bottle-neck or even a "show-stopper" in IC design. Five years ago, we at IBM Haifa Research Lab, suggested a powerful new design concept and methodology of using parameterized on-chip transmission line (T-line) devices from the early stages of the design flow in analog and mixed signal designs. This new IBM Haifa design concept is now implemented in all IBM SiGe technology design kits, propagates to all other IBM cutting-edge chip design technologies, and has been used for three years by customers outside and inside IBM. Four filed patents and eight IEEE papers have already been published by us on this subject. On December 2002, we received an IBM worldwide research accomplishment award for this project, as well as several personal prizes including an OIA (IBM Outstanding Innovation Award). We are now looking forward to seeing this become a new industry standard. Two independent research niches are intertwined in this fascinating subject: The first is ongoing research on a new (patented) comprehensive design methodology, which is based on the selective use of on-chip transmission line devices-lines with their own shielding, like "coax cables on chip"–as design components. This new concept for analog and mixed signal (AMS) design, is now being generalized to other design niches, such as ASIC and custom design, SOC design, and parts of mainstream VLSI design. It will be shown that properly shielded bends and junctions have very little effect compared with the straight T-lines effect, which highly simplifies the design methodology. The second is ongoing fundamental research on the science and art of signal propagation on interconnects on-chip, which utilizes and expands the over 20 year expertise gained by the IBM T. J. Watson Research Lab in the field of electromagnetic analysis. This second research niche has resulted in IBM Haifa's unique state-of-the-art transmission line simulation models, which work equally well in frequency and time domains, and are being constantly improved in Haifa. These high bandwidth models have been extensively verified by silicon hardware test-site VNA measurements from 50MHz till 110 GHz, as well as by several types of EM numerical analysis. These highly compact Haifa models are purely semi-analytical and explicit expressions, derived directly from Maxwell equations, passive by construction, and yield the correct asymptotic behavior in both the low and high frequency limits. The models cover both skin and proximity effects for microstrip T-lines, as well as a new full solution for the lossy silicon substrate effects for coplanar T-lines (signal attenuation, impedance frequency dependence, and slowing of wave propagation). The models are designed to cover the full bandwidth from DC until the cut-off frequency of the fastest transistors in the given chip technology. Document Segmentation with Application for the Book Publishing Industry (Abstract) Dan Chevion, IBM Haifa Labs I will describe a method for segmenting a scanned page into text, image, line-art, and background. Each segment undergoes specific image processing and compression routines, based on its type, and the document is then reassembled as in the original page. This procedure improves the print quality of the document, being as close as possible to the paper original, and eliminates artifacts that would otherwise result when printing a scanned document. Moreover, the disparate compression algorithms yield a reduced size file, improving performance in printers, servers, and networks. Automatic Event Handling in Uncertain Environments, with a Focus on Uncertain Event Inference (Abstract) Segev Wassekrug, IBM Haifa Labs In recent years, there has been an increased need for the use of active systems–systems required to act automatically based on events, i.e., changes in the environment. Such automatic actions can be either reactive (responding to past actual changes) or proactive (intended to prevent possible predicted changes). An emerging paradigm to enable such systems is the Real-Time Controller (RTC) –an ongoing control loop that consists of the following stages: Sense - monitoring the system Detect - detecting when an action is required Derive - collecting information from relevant data sources Decide - deciding on the optimal action to carry out Effect - carrying out the actual action in the system To implement such a RTC, advanced technologies are required. This presentation will introduce three such technologies: AMIT, an advanced complex event processing engine; ADI, middleware for modeling and monitoring entities and the dependencies between them; and ARAD, an optimization framework that enables optimization according to business objectives. The presentation will then focus on the major role uncertainty plays in such active systems, and I will describe an ongoing research project regarding event composition in the presence of uncertainty. Application Development Tools, with Focus on Support for Memory Use Optimization (Abstract) Ran Shaham, IBM Haifa Labs Excessive memory use in Java occurs when objects are no longer used, but remain referenced, and thus cannot be reclaimed. Excessive memory use has a negative impact on the performance and the RAS (Reliability, Availability, and Serviceability) of Java applications. In particular, our empirical studies show that around 40% of the space consumed by the application is no longer needed, yet this space cannot be collected by the garbage collector. Motivated by the above, we developed a tool called Muse, which points out code problems responsible for excessive memory use, and aids in the task of reducing the space consumption of a Java application. Initial experience with the tool shows an average space reduction of 15% for a set of Java benchmarks. In addition, we report our experience with application of the tool to large-scale, server-based software. Finally, we describe our static analysis algorithms for automating the task of space savings in Java applications. Model Based Software Testing (Abstract) Alan Hartman, IBM Haifa Labs Model based testing is still not a widely accepted industry practice despite the existence of academic and industrial case studies that discuss its advantages over traditional hand crafted testing practices. One of the main reasons for this is the lack of appropriate tools that scale to industrial sized systems. The Model Driven Testing (MDT) project at Haifa Research Labs is an attempt to remedy this. The MDT project has created a set of integrated tools for the behavioral modeling of distributed applications, test generation, test execution, and test analysis. Moreover the MDT tools are accompanied by a set of instructional materials and samples that provide an easy introduction to the methodology and tools used in model based testing. The talk will describe the MDT architecture for model based testing and discuss some of the ways in which the scalability problems have been addressed. Challenges with Advanced Development Projects (Abstract) Lior Leiba, IBM Haifa Labs Estimates are that more than two million people are working on more than 300,000 software projects in the US alone. Between one and two thirds of these projects will exceed their schedule and budget targets before they are delivered. A large portion (of the most expensive) of these projects will eventually be cancelled for being out of control. Many other projects are just left to wither on the vine or are abandoned by their sponsors or customers. This talk will give a short overview of the challenges in complex development projects and will present 'Rational Unified Process Best Practices' that will help to prevent your project from suffering from the consequences described above. The Rational Unified Process (invented by Rational, currently an IBM company) describes how to effectively deploy commercially proven approaches to software development for software development teams. These are called "best practices" because they are observed to be commonly used in industry by successful organizations. The talk will also cover other aspects of software development projects but will, of-course, not cover the long list of techniques, paradigms and methodologies that exists in the field. The Wonderful World of Web Search Engines (Abstract) Ronny Lempel, IBM Haifa Labs The World Wide Web is an ever-growing sea of information. The volume of data available online increases exponentially, as more and more people and entities publish data on the Web. In this chaotic arena of text, images, voice, and video, hundreds of millions of people from all over the world are surfing in search of information on every possible topic. Web search is made possible by complex information retrieval systems called search engines, which combine (among others) such disciplines as database technology, distributed computing and storage, statistical linguistics, and graph algorithms. Users submit queries to the engines, and expect their queries to be instantaneously answered with ranked lists of the most relevant URLs available online for each query. In order to meet these demands, search engines must collect and index billions of resources, and develop highly efficient retrieval and ranking algorithms, which are capable of effectively answering queries in almost a blink of the eye. This talk will highlight several challenges in search engine technology and Web structure research, such as applications of link analysis, the war against spam, e-commerce recommender systems, and power-law phenomena on the Web. Time permitting, some aspects of the business side of Web searching will also be covered. Experiments in TREC - the World Championship in Search (Abstract) David Carmel, IBM Haifa Labs Quality evaluation of search results is extremely important during the development of a search system. In this talk, I will overview TREC, the Text REtrieval Conference, which provides an infrastructure for large-scale experimentation, evaluation, and comparison of text retrieval methodologies. I'll then describe the search engine developed in our lab, Juru, and the experiments we were involved in TREC recently. Among them are: The effect of index pruning on search results Topic distillation using knowledge agents Merit estimation of search results Embedded Concatenative Text to Speech (Abstract) Ron Hoory, IBM Haifa Labs In recent years, major progress has been made in the quality and naturalness of text-to-speech (TTS) systems. Today, high quality speech can be produced by concatenative synthesis systems, where speech segments are selected from a large speech database and concatenated. To obtain high quality synthesized speech, a large amount of speech data is required. The large database size is required in order to cover as many phonetic contexts and as many acoustic environments as possible, and hence, avoid discontinuities at the concatenation points and reduce the amount of prosodic modifications. The talk will describe IBM's state-of-the-art concatenative text-to-speech system and the work done for reducing its footprint, to make it suitable for an embedded environment. Collaborating via the Cyber Space, Making Dear Ones Near Ones (Abstract) Natalia Marmasse, IBM Haifa Labs Communication between people who share the same physical space can be very rich. It is an interactive process, a collaborative act made up of both verbal and non-verbal cues. It is much more than simply a direct transfer of information. Today's telecommunication has several limitations. The telephone enables communication at a distance, in shared time, but without the richness we have when co-located. Its focus has been on verbal communication, restricting the non-verbal expression. Current communication devices lack back-channels helping us maintain awareness of those with whom we share communication space, they afford no way of inferring a person's situation before the communication has been initiated, and, moreover, they do not create opportunities for additional communication. WatchMe, embodied in a wristwatch, addresses mobile communication and awareness between people in a closely-knit group, e.g., family and friends. It aims to enhance the telecommunication between them by providing a relevant telepresence and a mobile platform facilitating various channels of verbal and non-verbal communication.