|
|
IBM Haifa Labs
Representative Achievements
| |
Active Middleware Technology (Amit)
Researchers in IBM Haifa's Active Management Technology group conceptualized and developed Amit, a cutting-edge tool that enables fast and efficient development of active functionality within applications and frameworks that are useful for e-business applications (such as stock market, business opportunities, and sales alerts), operations management, customer relationship management, command and control applications, and monitoring systems. Active functionality enables the application to detect customized situations without having to be aware of the occurrence of the basic events. Amit also features automatic inference of active behavior from high-level modeling concepts, such as business objectives and dependencies, and a proactive decision support framework. Amit was first deployed in IBM Global Services' e-business Management Service (ebMS) and is in the process of being implemented in several IBM software products. Amit is considered a major enabler for autonomic computing, both as a modeling tool for dynamic environments and as a detection engine for decisions on the run-time triggering of self-optimization and self-healing processes.
iSCSI
Today's e-business applications depend on storage solutions for high availability, high performance and superior scalability. Large-scale storage systems currently use the Fibre Channel standard to connect storage units and their multiple computers for data sharing, storage consolidation, and centralized management. Haifa researchers have developed iSCSI, a transport protocol for SCSI (Small Computer System Interface) that operates on top of the standard Internet Protocol. This groundbreaking solution is a convergence of storage connections and networking technology, providing competitive scalability for storage systems, with lower hardware, maintenance and operating costs. It also leverages the wealth of technologies developed for the Internet to the advantage of storage-area networks (SAN). iSCSI will undoubtedly accelerate the penetration of SAN to environments where Fibre Channel was seen as prohibitive.
Advanced Copy Services
Haifa's Storage Systems Technology department developed the advanced copy services for IBM's Shark Enterprise Storage Server. This high-end data storage system offers scalability, high speed, and high availability with no scheduled down time. IBM Haifa's copy functions are crucial to Shark's data management and disaster recovery capabilities. They include Concurrent Copy (allows reading a frozen, consistent copy of a volume, while keeping the original available for read and write commands); Flashcopy (creates an instant copy of a volume, while keeping both the source and target available for read and write commands); Peer to Peer Remote Copy (provides disaster recovery over medium distances by delivering a continuous, synchronous mirror of a volume to a remote location); and Extended Remote Copy (provides disaster recovery for large sets of data over long distances).
Juru Text Search
Haifa's Information Retrieval group is known for its innovations in the field of text search. The latest byproduct of this work is Juru - a full-text search engine written entirely in Java for text collections, ranging from Helpdesk FAQs to web portals, where precision is a must. Juru efficiently applies state-of-the-art as well as unique techniques to produce high-quality search results. Juru attained top results in the Web Track assessment of the National Institute of Standards and Technology's 2001 Text Retrieval Conference (see http://trec.nist.gov). It has been successfully integrated and/or deployed in several products and offerings, including IBM's Helpdesk solutions for self-help, domain specific search on the World Wide Web, a search engine for hand-held devices (PDAs), and IBM's WebSphere Portal Server.
Hebrew Speech Recognition
Haifa's Telecom and Media Systems group is developing the first technology that can recognize large vocabulary continuous speech in Hebrew. While there are other systems that recognize simple commands in Hebrew, this technology is in another league. It will enable users such as doctors and lawyers to dictate Hebrew documents instead of typing them. Other applications for this technology are advanced telephony-based services, audio mining in large audio/audiovisual databases, and assistance for disabled people. The main challenge for the Haifa team is to adapt IBM's general purpose speech recognition engine, ViaVoice, to face Hebrew-specific problems, such as the large number of inflections.
Model Checking Technologies and Formal Protocol Modeling and Analysis
Formal Methods (FM) for specifying and verifying complex computer chips and systems are increasingly in the forefront of software and hardware development. IBM Haifa's FM team has made a wide impact throughout industry and academia, and has become recognized as a world leader in the field, by moving formal verification of hardware from a theoretical concept to a viable technology. The contribution of the IBM Haifa team to the Formal Methods field is best represented by RuleBase - one of the most advanced formal verification tools available, which embodies Haifa's breakthrough Model Checking technology. RuleBase has been used in numerous ASIC and microprocessor development projects across IBM. It enables the timely launch of complex products through early assurance of functional correctness. Among the most advanced applications of RuleBase is Protocol Modeling and Analysis, in which an entire chip is modeled at the proper level of abstraction and validated against a set of high-level correctness criteria, prior to implementation. RuleBase is unique in its advanced state-space search algorithms, its repertoire of state-space reduction algorithms, and the size of hardware designs it can handle. In April 2002, the Accellera Formal Verification Technical Committee selected RuleBase's property specification language Sugar for IEEE standardization. As a result, the microelectronics and electronic design automation industries are expected to adopt Sugar as the specification language for their verification tools.
Advanced Test Program Generation Technology and Tools
Simulation of automatically generated test programs is the main means for verifying complex hardware designs. For more than a decade, IBM Haifa's Verification Technologies department has specialized in automatic Test Program Generation (TPG) technologies for high-end designs, with a focus on verifying processors, SMPs, and large systems. These technologies are recognized as the best in the field and are used throughout IBM's major hardware development labs. For many years, Haifa's flagship TPG tools were Genesys, which generates test programs for uni-processors, and Genie, which synthesizes concurrent test programs for multiprocessors. They provide a generic solution for various types of architectures by separating the generation expertise from the specific details of the design to be verified. Haifa's Genesys-Pro tool is the next generation technology for processor verification. The department has recently extended its range of advanced TPG technologies to include special-purpose, deep-knowledge test generators, such as FPgen (floating-point verification) and Piparazzi (micro-architecture flow verification). It is also involved in on-going research and development of a generalized solution for automatic system-level TPG, as embodied in X-Gen.
Model-based Automated Test Generation and Execution for Software
Haifa's Software Testing and Verification group has developed GOTCHA-TCBeans, a set of tools that assist testers to generate, execute, and organize function tests and system tests of software. The tools, whose underlying technology is an evolution of Formal Methods used for hardware verification, are particularly suited to testing Application Program Interfaces (APIs) and software protocols written in Java, C or C++. The toolkit supports reuse of many common testing functions, which saves testing time. Its systematic approach to test suite generation yields higher functional coverage, exposing more defects early in the software development cycle. It simplifies distributed application testing by enabling a single test to propagate transparently across many host computers. As a leader in the field, the Haifa team heads AGEDIS, a consortium of seven industrial and academic research centers in Europe and the Middle East researching the automation of software testing.
Constraint Satisfaction Techniques for Test Generation
A major challenge faced by Test Program Generation (TPG) tools is the efficient solution of the constraint satisfaction problems (CSPs) needed to generate high-quality, pseudo-random test programs. In this context, typical CSPs have various unique characteristics that render traditional CSP solution techniques inadequate to handle these problems properly. To meet this challenge, IBM Haifa's Verification Technologies department has developed a generic CSP engine, which allows users to model different CSPs that drive the generation of stimuli used for verifying new designs. This CSP engine uses random elements in an attempt to reach a different solution each time. It eliminates the need to create a multitude of solvers for specific problems. The CSP engine has been incorporated into Haifa's Generation Core technology, which provides algorithms and services that facilitate the fast introduction of new leading-edge TPG technologies, such as Genesys-Pro for processors, and X-Gen for systems.
Feedback Directed Program Restructuring (FDPR)
Researchers in Haifa's Code Optimization Technologies group developed FDPR, a feedback-based post-link optimization tool. The tool is suitable for optimizing very large programs or DLLs (dynamic link libraries). FDPR performs global optimization at the level of the entire executable, including statically linked library code. Since the executable to be optimized by FDPR will not be re-linked, the compiler and linker conventions need not be preserved, thus allowing aggressive optimizations not available to optimizing compilers. When used on large subsystems in a multiprocessing setting, FDPR provides significant performance improvements on software products running on IBM platforms. Part of this technology was embedded in the Parallel Tools Consortium's DPCL (Dynamic Probe Class Library) Project. FDPR has been incorporated into the AIX operating system version 5 and higher, and is also used internally within IBM on the Windows operating systems. In addition, prototypes were developed for Linux and S/390.
Development Resource Management Solution
Asset Locator is a low-cost solution for managing development resources in the enterprise, supporting search and reuse, collaboration, knowledge sharing, and impact analysis. It follows the Enterprise Development Resource Management (EDRM) paradigm, which defines the stages for enterprise-centric management of information assets, and how to leverage and integrate them in various tools and activities throughout the software lifecycle. Asset Locator uses a set of autonomous crawlers that scan enterprise servers/repositories to discover new development resources. A set of domain-specific analyzers process the discovered resources by identifying and extracting semantic and textual features. A powerful search engine enables clients to query the analyzed information. In addition, Asset Locator performs enterprise-wide impact analysis to improve development asset management. The design of Asset Locator as an extensible framework has enabled its easy integration into several IBM product offerings.
Garbage Collection
Haifa's Runtime Subsytems group has developed a series of scalable automatic memory managers, known as garbage collectors (GC), for Java on multiprocessor servers. A GC determines the memory objects that are no longer in use, collects their memory, and recycles it for future needs. In 1997 the Haifa group designed the first fully concurrent GC used in a commercial Java Virtual Machine (JVM), allowing Java code to operate pauselessly and take advantage of multiprocessor server machines. This GC was significantly more efficient than the GC that came with early JVMs, which was single-threaded and stopped all program activity while it ran. In 2001 the group completed the next generation of GC technology, which is parallel, incremental, and concurrent. It offers reduced overhead and low pause times for GC and makes GC feasible for the large heaps enabled by the new 64-bit architectures.
Clustering and High Availability
Haifa Researchers have made a key contribution to the architecture and design of the highly available iSeries cluster. They also implemented its heart - CLUE (CLUster Engine) - a group communication service providing a consistent membership view of cluster nodes or applications and the consistent exchange of messages. CLUE is a basic building block in enabling the single system image view and in maintaining a consistent shared state. The contribution to IBM revenues from this product is in the order of hundreds of millions of dollars. Clustering whole computers is a common cost effective solution enabling high availability, throughput, and scalability. Haifa researchers minimized the special coding required for applications to take advantage of a cluster infrastructure in a follow-on set of projects: highly available WebServer and Servlet Engine, Object Replication Service, Cluster Hash Table, ITRA and Easy technologies.
WebSphere Telecom Application Server (WTAS)
Researchers in Haifa's Service and CRM Technologies group designed this open development environment for telephony services with a highly scalable and flexible architecture. This environment enables the phasing out of proprietary hardware and software at the exchange. Service developers can use WTAS to create, customize, and configure their services without concern for protocol. Subscribers benefit by gaining a wider variety of services and the ability to subscribe and use services regardless of the brand of telephony equipment. WTAS will allow for the implementation of advanced services above and beyond the current 1-800 translation, follow me, or voice mail services. For example, with WTAS, subscribers will be able to store their calendar on a website and synchronize it with a WTAS reminder service.
InfiniBand Architecture
The Haifa Development Lab is an industry leader and pioneer of the emerging InfiniBand architecture (IBA). A series of three chips being developed in Haifa for IBM's InfiniBlue™ 4x series - InfiniBlue Target Channel Adapter (TCA), Host Channel Adapter (HCA) and Switch - met with unprecedented excitement when first introduced at industry exhibitions in November 2001 and February 2002. The InfiniBlue products provide an end-to-end solution designed to satisfy bandwidth-hungry network applications, such as those combining voice, data, and video on the Internet. Featuring high performance and reliability, these leading-edge devices provide ideal technology solutions for a range of network infrastructure components, including servers, SAN (storage-area-networks), and RAID (Redundant Arrays of Independent Disks) systems. The InfiniBand technology has a robust, layered design that improves how multiple computer systems and peripherals work together as a single high-performance and highly available server..
System on a Chip and Core Development
Haifa researchers were instrumental in the architecture, design and implementation of technology for integrating a logic system on a chip. The tools that were developed provide a common interface layer between communication interfaces and a common memory structure. This allows the rapid design of Application Specific Integrated Circuits (ASICs) by combining logic components (cores) designed elsewhere. Rapid turnaround is a key factor in the ASIC market. The first functioning chip, containing 11 logic cores, was created as part of this collaboration. Cores developed for this project include GbE MAC, 1394 MAC, Data Engine, and PCI.
General Parallel File System
Haifa's Storage and Systems Technology department was instrumental in developing IBM's first shared disk file system, the General Parallel File System (GPFS). This scalable, robust and high-performance file system assures data integrity and scalable performance under parallel reading and writing from many independent processes. GPFS supports file systems of several tens of terabytes and runs at I/O rates of several gigabytes per second. It can sustain normal operations in the face of multiple system and I/O failures. GPFS also enables users to group applications based on business needs rather than data location.
Recognition Compatible Voice Coder (RecoVC)
Researchers in the Telecom and Media Systems group developed the Recognition Compatible Voice Coder (RecoVC), a novel low bit rate speech coder/decoder (codec). RecoVC successfully avoids the degradation in recognition rates usually associated with low bit rate compression. It is intended for use in Distributed Speech Recognition (DSR) systems, where speech recording and recognition are performed at different locations or at different times, and where transmission channel bandwidth or storage space is limited. Examples of such systems include Web-based voice portals, such as voice mail and Interactive Voice Response (IVR) services accessed from Internet phones, cellular phones, or other portable devices, as well as Personal Digital Assistant (PDA) devices that record voices for later recognition. Other speech database services with advanced features such as word spotting, speaker identification, and speech transcription can also benefit from the RecoVC technology.
Integrated Digital Medical Records (IDMR)
IDMR is a hospital-wide Picture Archival and Communication System (PACS) that provides a full solution for film-less and paper-less radiology. Digital images from all types of modalities are stored in a long-term hierarchical storage system for immediate access by hundreds of workstations. The entire radiological workflow is managed by IDMR, including voice dictation and text reporting. IDMR conforms to the DICOM and HL7 standards and is FDA approved. The system was developed in cooperation with the 2000-bed Sheba Hospital in Tel-Hashomer, Israel, where it archives 4TB of image data annually. In the Haifa Lab, IDMR paved the way to new healthcare oriented activities, such as the medical Hebrew speech recognition pilot; a powerful general Medical Storage System (MSS); and Integrated Medical Records (IMR). IMR is middleware that integrates and correlates medical records from diverse sources for secure and privacy-protected sharing of patient information among different and disparate organizations.
State Hermitage Museum Project
This IBM Corporate Community Relations project with the State Hermitage Museum in St. Petersburg, Russia, began in January 1997. It aimed to transform how people around the world experience the Hermitage Museum and its collections. As part of this project, Haifa's Solutions Technology group developed Visitor Information Kiosks that provide museum visitors with dynamic multimedia navigational tools and information on the museum. The Java-based application, created by Haifa researchers, was the first to use Java technology in an IBM kiosk application. The Hermitage kiosk has been installed in many museums around the world, including the UK, Japan, Brazil, Peru, Netherlands, South Africa, Finland and others. The Hermitage website, which features a searchable digital library of high-resolution images from 12 categories of art, uses "Zoom View," an innovative Java-based technology developed in Haifa especially for this project.
Trust Establishment
The Internet enables connectivity between many entities that don't know each other, but to enable e-business between them, the parties need to establish some level of trust. Researchers in Haifa's Trusted e-Business group developed an innovative approach that enables businesses to define flexible policies for mapping strangers to predefined business roles, based on presentation of third-party certificates. Issuers of the certificates need not be known in advance, as long as they provide certificates identifying them as trusted authorities, according to the policy. This allows bottom-up, grassroots creation of trusted issuers, hence enabling e-business. This approach extends existing role-based access control mechanisms by providing automated role assignment.
Optimized Airline Crew Scheduling
Haifa researchers in the Software and Services department developed a highly successful crew scheduling and assignment system for commercial airlines, with El Al Israel Airlines as the prime customer. ECO.2000 solves the problems of scheduling cockpit and cabin crew, while minimizing operational costs and balancing workloads among individual crew members. It is based on advanced technologies and research, with additional innovations developed in Haifa. El Al airlines found that ECO.2000's deep computing solution - combining mathematical modeling, computation power, and heuristics - significantly improved the crew planning process, resulting in a measurable decrease in crew related costs and better crew utilization.
Network Design and Analysis (NetDA)
Haifa researchers developed NetDA/2, a powerful OS/2-based tool for the design, analysis and evaluation of wide area communication networks. It supports most communication architectures and protocols. Via a friendly interface, which employs geographical background maps and graphical network topology layouts, the user specifies the expected communication load, required Quality of Service (QoS), and infrastructure costs. Based on this input, and using advanced data structures and algorithms in graph theory, the tool designs an entire network, optimized for cost, performance, and reliability. Applying analysis and simulation, NetDA/2 also allows planners to manage changes to the network prior to actual implementation.
Linear Sorting Algorithm
Haifa researchers were innovative in developing IBM's SMARTSort utility for AIX in 1995. SMARTSort was significantly more efficient than the previous AIX sort utility. It is a sort by interval algorithm that positions the intervals dynamically, based on the outcome of a statistical sampling, such that for the vast majority of files, CPU time grows only linearly with the size of the file. This sorting algorithm was invented and patented by IBM Haifa and it was innovative as the first commercial implementation of a linear sort. SMARTSort was also implemented in OS/2 and Windows NT, and supported multiple data types, including double byte characters for national language support, bit level sorting and merging.
Automated Parcel Sorting
Manually sorting parcel mail for delivery is a costly and time-consuming process. To address this problem, Haifa researchers, together with IBM's Almaden Research Center and Postal Solutions Center, developed an automated parcel sorting solution ("ParcelVision"). This solution involves scanning the mail as it moves along a conveyor belt, locating and reading bar-codes, analyzing parcel image layout, locating and enhancing the destination address, reading the handwritten or printed address, and encoding the address to a full zip code. Haifa developed several novel algorithms, image processing tools and reliable parallel processing applications for these projects. With the help of Haifa's contributions, IBM won the first two contracts to be awarded anywhere for parcel automation, at German Post and Swiss Post.
Forms Processing and OCR
As a pioneer in the field of forms processing, IBM Haifa developed a comprehensive solution for creating highly compressed forms archives and virtually error-free data entry. Technologies developed at Haifa allow reliable form recognition, form dropout, character recognition, logical text analysis, and tools for effective data validation and correction. These systems have been used by the Israeli and Swiss population censuses and by US state tax agencies.
Compiler Optimization
Over the years, Haifa researchers have conducted innovative work with compiler optimization techniques, enhancing the software performance of a number of IBM products and establishing the Haifa Lab as a leader in this arena. Optimization techniques include: register allocation, instruction scheduling, software pipelining, collapsing data dependencies, exploitation of machine specific resources and I-cache, and special instructions optimization. These technologies have made significant improvements to a number of IBM products, including the XLC compiler for AIX and the OX compiler for AS/400. Current activity in the Optimization group includes an improved JIT compiler for the Java environment, a C compiler for a new DSP architecture with competitive performance results, and investigation of techniques to reduce the power consumption of running applications.
Java Static Analysis
Over the past few years, IBM Haifa has been developing an infrastructure for static analysis of Java bytecode. Our motivation is to support program understanding, code evaluation, testing, and optimization. The infrastructure includes the development of analysis techniques, and the production of certain tools that exploit these techniques. The development code base consists of a set of libraries, such as CFParse for parsing and editing a single classfile; JAN for analyzing a whole Java component and focusing on global representations, such as the component's call graph; and certain generic data-flow engines. We implemented several utilities upon this set of libraries, including TOAD, an umbrella framework for a set of core libraries and tools that monitor, understand, and optimize Java applications; JAMA, which detects immutable fields and classes; Struts Checker, which collects information about Java bytecode in Struts-based projects; and JavaDoc Enhancer, which enhances the JavaDoc documentation with valuable semantic information.
BookManager
BookManager was a softcopy publishing and presentation solution based on GML (General Markup Language) developed at IBM from 1985-1993. The development of BookManager drew upon Haifa's experience in specific technologies, including information retrieval, user-interfaces, and information presentation. This was one of the first projects to demonstrate Haifa's ability to play a major role in the development of a project in different Operating System environments (OS/2, AIX, OS/400), as well as in project testing, managing sub-contractors, documentation, and maintenance. This project also served as a research vehicle on the subjects of hypertext and information retrieval, which resulted in numerous papers, patents, and awards. The project saved IBM millions of dollars in fulfillment costs (printing) and received the Green Seal of Approval as an environmentally responsible product.
DASD I/O Simulator
The DAIOS project, a DASD (Direct Access Storage Device) I/O Simulator, was a first for IBM Haifa in many ways. It paved the way for Haifa's involvement in storage technology, which now plays a major role in Haifa's research. It was the Haifa Scientific Center's first major project for an IBM development lab in the United States. The DAIOS simulator was used to test and tune the largest mainframe computers developed by IBM Poughkeepsie in the 1980s. It created significant savings for the lab, in terms of time, money and floor space, since developers could test these large machines without dedicating hundreds of disks to the testing process. By eliminating interruptions caused by mechanical disk failure, DAIOS made benchmark testing far more efficient. With its use of sophisticated software written in microcode and running only on a controller, DAIOS was the forerunner of Haifa's Advanced Copy Services developed for the IBM Shark Enterprise Storage Server.
Machine Assisted Translation
Researchers in Haifa's Natural Languages Department led an international research and development effort carried out in several IBM centers in Europe and the US. The project tackled different aspects of automatic computerized translation from one language to another. Haifa's responsibilities included technical leadership of the project, incorporation of statistical techniques, interactive disambiguation, integration of "translation engines," and development of debugging aids. Haifa also developed an English-Hebrew translation prototype which was used to demonstrate the project's achievements.
Correctly Rounding Math Library for Java
Researchers in IBM Haifa's Software and Services department created the IBM Accurate Portable Mathematical Library (IBM APMathLib) - a package of fast and extremely accurate math routines that compute 12 of the standard common transcendental functions. Unlike standard library routines, the computed results are the exact theoretical values correctly rounded to the closest number representable by the IEEE 754 double format. If the theoretical result is equally distant from two neighboring IEEE 754 values, the functions return the "even" of the two. The main advantage of this collection is that its computations are bit-by-bit compatible among the supported platforms, enabling true Java portability.
Signal Processing for the Hearing Impaired
IBM Haifa initiated this project as a contribution to the International Year of the Disabled in 1980. The project demonstrated the applicability of advanced signal processing techniques to filter out background noises, thus increasing speech intelligibility for hearing aid users. To alleviate the "cocktail party" syndrome, the system uses a signal processor to separate the desired speaker from competing noise, utilizing the fact that the desired speaker and interfering noise sources are at different spatial locations. An evaluation of the system, conducted by Tel Aviv University's Department of Communications Disorders, indicated that the system could contribute markedly to the functioning of a hearing impaired person in a noisy environment. Algorithms developed for this project contributed to Haifa's leading edge in speech recognition technology.
Computer Aided Processing of Medical Ultrasound Data
The goal of this joint effort between IBM Israel/Sheba Hospital in Israel and IBM Austria/ University Hospital in Vienna was to improve the diagnostic capabilities of medical ultrasound technology through the application of new signal and image processing techniques. This project was launched in 1977, when ultrasound scanning was just beginning to expand its technological horizons. The project involved real-time data acquisition of log compressed RF signals, followed by analytical signal computation and additional compression; resolution enhancement by single scan deconvolution; spectral feature mapping based on processing of RF signals; and filtering of video signal for speckle reduction. Haifa developed novel techniques for transforming ultrasonic images into the corresponding tissue attenuation images. These images, in turn, advanced diagnostic capabilities -for diffused liver diseases, in particular. This technology was licensed to a medical instrumentation company. Haifa also developed compression technology for ultrasound images that contributed to the ability to build ultrasound image databases.
Computer Models for Agricultural Farm Management
One of the Scientific Center's first joint projects with an IBM center outside of Israel, its goal was to improve the efficiency of agricultural production. This project included the development of a special language for formulating linear programming problems and an interactive system for constructing irrigation time tables using an innovative, efficient hydraulic network solver. The latter was implemented in diverse agricultural frameworks, including Israeli agricultural cooperatives (i.e., kibbutzim and moshavim). It was also used in the planning of urban and regional water supply systems. In 1981, this technology received the Information Processing Association of Israel (IPA) Award.
|
|
|
