Software for the ACP multiprocessor system

Biel, J.; Areti, H.; Atac, R.; Cook, A.; Fischler, M.; Gaines, I.; Kaliher, C.; Hance, R.; Husby, D.; Nash, T.; Zmuda, T.

doi:10.1016/0010-4655(87)90170-6

Cited by 9 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the first implementations of event-oriented multiprocessing for particle physics [5,6], the multiprocessor management software had to do a great deal of hard work. Clients frequently had no significant operating system, primarily because memory was expensive.…”

Section: The Multiprocessor Manager's Tasksmentioning

confidence: 99%

See 1 more Smart Citation

A simple multiprocessor management system for event-parallel computing

Bracker

Gounder

Hendrix

et al. 1996

IEEE Trans. Nucl. Sci.

View full text Add to dashboard Cite

Offline software using TCP/IP sockets to distribute particle physics events to multiple UNIX/RISC workstations is described. A modular, building block approach was taken, which allowed tailoring to solve specific tasks efficiently and simply as they arose. The modest, initial cost was having to learn about sockets for interprocess communication. This multiprocessor management software has been used to control the reconstruction of eight billion raw data events from Fermilab Experiment E791. The E791 Reconstruction TaskFermilab Experiment E791 accumulated a large dataset (50 Terabytes, 20 billion events, 24 000 8mm Exabyte tapes) in 1991 and early 1992 [1,2]. As might be expected, the reconstruction and analysis of this data challenged available computing resources; event reconstruction alone required over 10 000 mips-years of processing power. For 2 1 2 years, reconstruction processing was underway at four different locations [3]. The three largest sites used clusters or farms of commercial UNIX/RISC workstations connected together by thin-wire Ethernet. Within each farm, many processors operated together; data management and system control were exerted from a single point via multiprocessor management software. Here we describe the multiprocessor management software developed and used at the University of Mississippi [4]. The Mississippi farm hardware is shown in Figs. 1 and 2.Most of the large-scale computing needs encountered in particle physics -reconstruction and analysis of recorded data and generation of simulated data -are event-oriented. Each event's data packet is extracted from an input stream and processed in isolation from other events. The results from each event's analysis are merged into an output stream. The computing power required to process each data packet is significant relative to the time needed to transport the data, even over datapaths of modest throughput. Many other scientific computing problems (e.g. small-object recognition in astronomical images) conform to the same model. In all such problems, it is trivial to divide the total computing task over many processors; each processor is simply given a share of the events to process independently of its peers.Most reconstruction and simulation tasks are developed and tested as single-processor programs. They can be outlined in the following manner: initialize data structures initialize input and output data streams read run startup information from input data stream read and store run calibration constants for every event from the input stream... unpack and check the input data PROCESS THE EVENT (most of the work) for good events, pack the processed event into output stream produce a report characterizing the processing job stop 1 When the program is "nearly working," it is then moved into the multiprocessing environment for final testing and production. If it has been written in a sensible fashion, with dataflow management tasks cleanly separated from event processing tasks, then the transition from single processor to multiproc...

show abstract

Section: The Multiprocessor Manager's Tasksmentioning

confidence: 99%

“…In the early days of "computer farming" in particle physics, there were not suitable commercial processors available, so we built our own [5,6]. After a short time, we were put out of the processor building business by high-powered workstations offered by several vendors [17].…”

Section: Future Directionsmentioning

confidence: 99%

A simple multiprocessor management system for event-parallel computing

Bracker

Gounder

Hendrix

et al. 1996

IEEE Trans. Nucl. Sci.

View full text Add to dashboard Cite

show abstract

“…These farms generally do not have operating systems running on the satellite nodes, but rather very primitive kernels sufficient to execute FORTRAN programs. For example, this kernel is several hundred lines of FORTRAN on the Fermi National Accelerator Laboratory Advanced Computer Program (ACP) [18]. Although this is a simple approach that performs well, it can be problematic for the development and debugging of applications.…”

Section: Introductionmentioning

confidence: 99%

The Parallel Processing Compute Server

et al. 1991

View full text Add to dashboard Cite

The Parallel Processing Compute Server (PPCS) is a distributed-memory multiprocessing system consisting of System/370™ microprocessors (33 at present) interconnected through a matrix switch. This paper describes the hardware configuration, the extensions to the System/370 instruction set that are provided to support the distributed memory and interprocessor signaling, the modifications to the VIUI/SP operating system that allow it to run effectively on many closely coupled processors (most of which have no disks), and the application-support layer, which permits FORTRAN programs to take advantage of the highly parallel environment. Development of the PPCS is a joint effort of the IBIVI Boblingen Development Laboratory and the IBM Thomas J. Watson Research Center. A prototype PPCS has been installed at CERN. IntroductionThe Parallel Processing Compute Server (PPCS) is a distributed-memory multiprocessor that uses message passing for interprocessor communication. It comprises an IBM Enterprise System/9373 (ES/9373) Model 30' [1] 1 The ES/9373 Model 30 is an entry-level processor of the Enterprise System/QSTO™ (ES/9370™), which is a family of mM System/370™ processors that support many users and applications in the commercial, engineering, scientific, and industrial environments. integrated host (I-host) system and 32 (at present) System/370 satellite processors [2], all interconnected via a matrix switch designed to accommodate 62 ports. The I-host is the only processor with I/O capability; otherwise, the satellite processors are functionally identical to the I-host processor. Running under the control of Parallel VM, a modified form of the IBM Virtual Machine/System Product (VM/SP) operating system [3], the PPCS is suitable for a variety of numerically intensive computing programs.Parallel VM features "diskless" operation on the satellite processors (but is designed to exploit I/0-capable satellite processors, if they exist), fast interprocessor communications, local and remote inter-virtual-machine message-passing, and, to a large degree, a single-system image as seen by the application programmer.Distributed VS FORTRAN, a prototype programming environment for PPCS, has been developed. VS FORTRAN [4] application programs must be reorganized by the user to exploit parallel processing by originating, scheduling, and synchronizing subtasks. A subroutine library for task management and data transfer has been developed in the spirit of IBM Parallel FORTRAN [5]. The implementation of this Distributed VS FORTRAN library is based on a CMS [3] extension, called CS/X, that takes advantage of the Parallel VM functions.A PPCS with 32 satellite processors. Parallel VM, and CS/X has been installed at the European Organization for Nuclear Research (CERN) and is running applications used by the high-energy physics (HEP) community. At CERN, ®Copyright 1991 by International Business Machines Corporation. Copying in printed form for private use is permitted without payment of royalty provided that (1) each reproduction is done w...

show abstract

Multiprocessors for high energy physics

Pohl¹

1987

Computer Physics Communications

View full text Add to dashboard Cite

Software for the ACP multiprocessor system

Cited by 9 publications

References 0 publications

A simple multiprocessor management system for event-parallel computing

A simple multiprocessor management system for event-parallel computing

The Parallel Processing Compute Server

Multiprocessors for high energy physics

Contact Info

Product

Resources

About