Computing

views updated

COMPUTING

Applications relevant to elementary particle and high-energy physics (HEP) computing can be categorized as follows:

Triggering and data acquisition
Simulation
Data handling and storage
Commodity hardware and software
Data analysis and visualization
Control and monitoring systems
Information systems and multimedia

Triggering and Data Acquisition

In addition to their specialized detection and measurement systems (for example, calorimeters, drift chambers, etc.), the detectors in high-energy physics experiments are, in fact, sophisticated computing systems. Without reliable triggering and data acquisition (DAQ) computing systems present in these detectors, all other experimental computing is of little consequence. Triggering and DAQ systems ensure that the physics events occurring in the detector are observed, measured, and accurately transformed into analyzable data. In a typical experiment, the first level trigger, implemented in hardware, initiates the data collection. Data from the front-end electronics are digitized and collected with electronic data modules. A readout computer reads experiment conditions from the control system, reads event fragments from the data modules over the local network, and builds events from fragments. These event data are then written to buffer storage and/or transmitted via a local network to archival storage for additional processing and eventual analysis.

The scale of triggering and DAQ systems can be seen in the design of the ALICE experiment at the Large Hadron Collider (LHC) at the European Laboratory for Particle Physics (CERN) in Geneva, Switzerland. The ALICE detector will measure up to 20,000 particles in a single interaction event resulting in a data collection rate of approximately seventy five million bytes per event. The event rate is limited by the bandwidth of the data storage system. Higher rates are possible by selecting interesting events and subevents or efficient data compression.

Simulation

A computer simulation (sometimes referred to as a Monte Carlo simulation) of particle interactions in the experimental configuration is essential to most HEP experiments. Computer software providing these simulations plays a fundamental role in the design of detectors and shielding components, in the investigations of the physics capabilities of the experiment, and in the evaluation of background (nonexperimental, for example, cosmic and/or terrestrial radiation) data. Simulation software must be complete and capable of generating simulated experimental data comparable in scope to genuine experimental data. The simulation software must support a description of the experimental detector from the point of view of the materials used and the geometry adopted, both for the structural and the active event-detecting components. The configurations adopted for the data output and the logic of DAQ on the quality of the physics results are also modeled in order to evaluate their impact on the overall performance of the detector. The simulation must be able to describe the properties and the physics processes of the particles involved both in the expected signal/output and in the background. Especially important is the capability to handle physics processes across a wide energy range, which in such experiments simulation may span several orders of magnitude. An ideal simulation system is also flexible and open to evolution and to the integration of external tools. This is particularly important since a number of software tools are already commonly used within the scientific community where a particular experimental environment may require the ability to extend the simulation functionalities, for instance, to include the ability to deal with peculiar physical processes. One of the most powerful and widely used simulation toolkits is GEANT4 developed at CERN.

Data Handling and Storage

Particle physics experiments generate enormous amounts of data. For example, the BaBar experiment at the Stanford Linear Accelerator Center (SLAC) is designed to accommodate 200 terabytes (200 million million bytes) of data per year. As of April 2002, the BaBar database contained over 500 terabytes of data in approximately 290,000 files. This database is the largest known in the world (with the possible exception of some with military/government content). Such data rates and database sizes push the limits of state-of-the-art data handling and database technologies.

In order to handle such large volumes of data, experiment data handling and storage systems/database must be designed to

provide reliable and robust storage of the raw detector data, simulation data, and other derived data;
keep up with production processing; be able to process raw data files within minutes of writing them to tape;
provide easy, rapid, and intuitive access to data on a variety of systems at a wide variety of locations where processing and data storage resources are available to physicists;
provide accurate detailed information on the processing steps that transformed event data— from the trigger through reconstruction and all the way to the creation of individual or group datasets;
provide mechanisms for policy-based allocation and use of disk, central processing unit (CPU), network, and tape drive resources.

Commodity Hardware and Software

Commodity hardware and software refers to the hardware and software architectures and configurations used to accomplish off-line batch and interactive data processing. In the past such processing was often accomplished by large mainframe computers. In recent years, large (200 or more) compute farms of inexpensive computers have become a common replacement for these mainframe systems. These compute farms are fundamentally groups of networked desktop systems (without monitors and keyboards) that are housed in a single location and which function as a single entity. A computer farm streamlines internal processes by distributing the workload between the individual components of the farm and expedites computing processes by harnessing the power of multiple CPUs. The farms rely on load-balancing software that accomplishes such tasks as tracking demand for processing power from different machines, prioritizing the tasks, and scheduling and rescheduling them depending on priority and demand that users put on the network. When one computer in the farm fails, another can step in as a backup. Combining servers and processing power into a single entity has been relatively common for many years in research and academic institutions. Compute farms provide an effective mechanism for handling the enormous amount of computerization of tasks and services that HEP experiments require. Farms of Intel-based computers running the Linux operating system (OS) have become common at many HEP institutions.

The computing grid is the next generation of compute farms. A grid is a distributed system of computing resources (a cyberinfrastructure) in which computers, processor farms, disks, major databases, software, information, collaborative tools, and people are linked by a high-speed network. The term grid was coined as a result of the analogy with an electrical power distribution system. Grid resources are made available transparently to a distributed community of users through a set of new middleware that facilitates distributed collaborative working in new ways. The nine-institution Particle Physics Data Grid collaboration—consisting of Fermi National Laboratory, SLAC, Lawrence Berkeley Laboratory, Argonne National Laboratory, Brookhaven National Laboratory, Jefferson National Laboratory, CalTech, the University of Wisconsin, and the University of California at San Diego—will develop the distributed computing concept for particle physics experiments at the major U.S. high-energy physics research facilities.

Data Analysis and Visualization

Analysis systems are often at the core of an experiment's physics efforts, and the constraints imposed by those systems can heavily influence the physics event reconstruction and analysis framework. Conversely, an analysis system which lacks key features (or worse, implements them incorrectly) can be a serious handicap. Physicists are constantly searching for new and interesting ways to extract physical information through two-dimensional and three-dimensional computer visualization/modeling, animation, histogram plotting, etc. Key also is the development of techniques for data interactivity— methods for interacting with a program or data. These techniques often include graphical user interfaces (GUIs) but also scripting, browsing and other technologies. There have even been some attempts to utilize virtual reality techniques wherein a physicist becomes "immersed" in experimental data. Development of data analysis and visualization tools has been the subject of numerous international collaborations. The result has been the creation of specialized software libraries used, supported, and maintained by these collaborations but generally available to all physicists.

Control and Monitoring Systems

The infrastructure surrounding experiment detectors is highly complex. The hardware devices used in detectors and the systems of experiments consist of commercial devices used in industry, specific devices used in physics experiments, and custom devices designed for unique application. The control and monitoring system must insure that these devices interface correctly with one another by providing testing and error diagnostic functionality. The administrative component of a control and monitoring system provides access to the control of an experiment often distributed between supervision and process control functions.

Information Systems and Multimedia

The World Wide Web (WWW) is the best example of how the requirements of physics research and the need for experimental collaboration have led to developments in information systems and multimedia. The Web was developed to allow physicists in international collaborations to access data and information easily, quickly, and in a device-independent (i.e., computer and operating system) manner. There has been increasing use of collaborative environments supporting point-to-point and multipoint videoconferencing, document, and application sharing across both local and wide area networks; video on demand (broadcast and playback); and interactive text facilities. Resources such as the HEP preprints database at SLAC and the Los Alamos National Laboratory electronic preprint server, officially known as the e-Print Archive, support physicist research and authoring. The first U.S. web server, at SLAC, was installed to provide access to the pre-prints database. Other information systems and multimedia applications include electronic logbooks used to improve and replace paper logbooks, and streaming media servers to provide widespread access to seminars and lectures.

See also:Detectors, Collider