P. A. Buhr and R. A. Stroobosscher, `The System: providing light-weight concurrency on sharedmemory multiprocessor computers running UNIX', Software---Practice and Experience, 20, 929--963. (1990).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....kernel with automatic garbage collection and supports all fundamental objects and methods of computer algebra. Several algebraic applications have been implemented in this library such as the Partial Cylindric Algebraic Decomposition Method and the Grobner Bases Algorithm. ffl The System [3] is a library that supports light weight concurrency on shared memory multiprocessors and Unix workstations. The System kernel distributes tasks (lightweight processes) among virtual processors which are Unix processes scheduled by the operating system. On a multi processor, virtual processors are ....

Peter A. Buhr and Richard A. Stroobosscher. The System: Providing Light-weight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. Software --- Practice and Experience, 20(9):929--964, September 1990.

....A precise syntax has been defined to describe all these objects,providing constructs to describe both their internal and external behaviour. The adopted syntax has borrowed much from that used in Concurrent C [30] although it is not identical to it. Some elements of the so called System [31] have also influenced our design. Unfortunately, none of these tools (or others) could be used directly to implement our concepts, hence the necessity to design our own language. Again, description of these syntactic and semantic details would go beyond the scope of this paper. 5. General ....

P. Buhr and R. Stroobosscher, "The System: Providing light--weight concurrency on shared-- memory multiprocessor computers running UNIX," Software --- Practice and Experience, vol. 20, pp. 929--964, 1990.

....If two nodes attempts to pin the same pages and each node only pins some of the pages, then the system will deadlock. This solution is crude, but effective. 2.4 Related Work Many threads packages have been written. Those that support efficient parallelism include Threads [21] Presto [6] System [9], C [8] and Sun Lightweight Processes [51] We will call these standard packages because they have a stack for each thread. The goal of standard packages is to provide the user with a natural thread abstraction and many of the usual concurrent programming primitives; different packages provide ....

Peter A. Buhr and R.A. Stroobosscher. The System: Providing light-weight concurrency on shared memory multiprocessor computers running UNIX. Software Practice and Experience, pages 929--964, September 1990.

.... of a parallel computer algebra package [SH93] and of a parallel functional programming language [Sch96] RT owes to PACLIB its central ideas of thread scheduling [Sch94] which are based on virtual threads [KW92] respectively lazy threads [MKHJ91] PACLIB was built upon the System package [BS90]; the contents of this thread package taught us the essential concepts of multi threading. The RT assembly code for mutual exclusion was mainly taken from the System respectively its successor C [BDS92] RT considerably owes to the development group of the symbolic computation package ....

Peter A. Buhr and Richard A. Stroobosscher. The System: Providing Light-Weight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. Software --- Practice and Experience, 20(9):929--963, September 1990. System available at ftp://plg.uwaterloo.ca /pub /uSystem /.

....net of computers which provides coarse grain parallelism. In this case it would be possible to perform a detailed analysis of various schemes of parallelism for different computational problems. Furthermore such an implementation could be compared with other 11 existing libraries like the System [BS90]. Finally we could combine our work on the implementation of particular algorithms from CA in SML outlined in [GL96] with the approach presented here. ....

P.A. Buhr and R.A. Stroobosscher. The System: Providing Light-Weight Concurrency on SharedMemory Multiprocessor Computers Running UNIX. Software - Practice and Experience, 20(9):929--963, September 1990.

.... is the concept of virtual PEs the difference being that the different code fragments can interact with each other with the mapping to physical PEs hidden in the runtime system [581] Examples include the multiprocessor implementation of concurrent C [124, 224] and various thread packages [130, 612, 59, 192, 87, 105, 55, 529]. The Psyche operating system on the BBN Butterfly provides a similar notion of virtual PEs at the operating system interface [510] Scheduling in Mach can also be interpreted in this light, with threads as virtual PEs that are executed by whatever number of physical PEs are assigned to the ....

....has to be made (right of figure) A different approach altogether is to forgo all information exchange, and strive to approximate the optimal efficient behavior. For example, in the System, separate threads are designated to perform blocking I O operations on behalf of those running user code [87]. With cooperative threads, there is an option to create or schedule a new kernel thread whenever an existing one blocks [286, 218] This prevents PEs from going idle if a kernel thread blocks when performing a blocking operation on behalf of a user thread. MAXI envelopes take this a step further, ....

P. A. Buhr and R. A. Stroobosscher, "The system: providing light-weight concurrency on shared-memory multiprocessor computers running UNIX ". Software --- Pract. & Exp. 20(9), pp. 929--964, Sep 1990.

....rather misleadingly called monitors. However, notify and wait operations must be explicitly invoked within such sections, rather than being automatically associated with entry and exit. There are many other thread packages available providing lightweight processes with shared memory communication [40, 41, 87, 155]. Rendezvous. Rendezvous based programming models are distributed memory paradigms using a particular cooperation mechanism. In the rendezvous communication model, an interaction between two processes A and B takes place when A calls an entry of B, and B executes an accept for that 38 entry. An ....

Peter A. Buhr and Richard A. Stroobosscher. The System: Providing light-weight concurrency on shared-memory multiprocessor computers running UNIX. Software Practice and Experience, 20(9):929--963, September 1990.

....drawback to this approach is that POSIX was designed as an application program interface, with features such as real time scheduling support that may add overhead for parallel systems. A lowerlevel interface designed specifically as a compiler target would most likely result in better performance [2, 9] and will be investigated in future research. To summarize, the mapping of computation to physical processors is determined by both the mapping of threads to contexts and the mapping of contexts to nodes. The relationship between nodes, contexts, and threads is illustrated in Fig. 1. Global ....

Peter Buhr and R. Stroobosscher. The ¯system: Providing light-weight concurrency on shared-memory multiprocessor systems running Unix. Software Practice and Experience, pages 929--964, September 1990.

....This model is however of much higher abstraction than message passing. But the function to execute as request and the reply function have no clear connection. In contrast, on machines with shared memory there is the well known fork join parallel programming model used by most threads packages [GGM93, Mue93, BS90]. This paper describes a system which extends this programming model used on shared memory threads to a distributed memory environment. It therefore combines ease of use with scalability and can be used on both shared and distributed memory machines. Chapter 2 describes the abstract properties of ....

Peter A. Buhr and Richard A. Stroobosscher. The ¯System: Providing LightWeight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. Software-Practice and Experience, 20(9):929--963, september 1990.

....polynomial gcd and resultant computation, polynomial factorization, real root calculation and algebraic number arithmetic. Several algebraic applications have been implemented in this library such as the Partial Cylindric Algebraic Decomposition Method and the Grobner Bases Algorithm. ffl System [3] is a library of C functions that supports concurrency on shared memory multiprocessors and Unix workstations. The System kernel distributes tasks (light weight processes) among virtual processors which are Unix processes scheduled by the operating system. On a multi processor, virtual ....

P. Buhr and R. Stroobosscher. The ¯System: Providing Light-weight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. Software --- Practice and Experience, 20(9):929-- 964, September 1990.

.... Delta Delta Delta request oe reply (b) The server waits Figure 3.3: Request reply rendezvous cannot accept other requests; thus, calls are necessarily mutually exclusive. Some languages, such as Ada [DoD83] and Concurrent C [GR86] as well as concurrency libraries such as the System [BS90] use this style of bidirectional message passing as their communication 7 The author s personal experience backs this up. An early version of the Pegasus system [RG86] used asynchronous message passing, but we had great difficulty in debugging our programs. Our experience with the ....

....from copying collection. The collector also suffers from the problem of poor real time responsiveness. 11.3 Comparison with the System To put these measurements into perspective, I implemented a similar set of benchmarks in version 4. 4 of the System, which is a C light weight process library [BS90] The System provides threads and a request reply communication primitive (it also has shared memory primitives) but it does not have selective communication. It runs on the SPARCstation and DECstation, but not on the NeXT. Table 11.5 reports the results for the SPARCstation Table 11.5: ....

Buhr, P. A. and R. A. Stroobosscher. The ¯System: Providing light-weight concurrency on shared-memory multiprocessor computers running UNIX. Software -- Practice and Experience, 20(9), September 1990, pp. 929--963.

....1. Saclib [3] is a large library of algebraic algorithms written in the C language. These algorithms are based on a list processing kernel that provides automatic garbage collection. Furthermore, the package supports exact arithmetic with arbitrary precision integers and rationals. 2. The uSystem [1] is a package that provides light weight concurrency on monoprocessor workstations and shared memory multiprocessors running under the UNIX operating system. The System provides an efficient management of tasks i.e. light weight parallel processes. Sponsored by the Austrian Science Foundation ....

....but we believe that the main features have already stabilized. The system has been implemented on a Sequent Symmetry shared memory computer with 20 processors running under the Dynix ptx version of UNIX System V.3. However, it is portable to any other machine supported by the System package [1]. This short paper just gives an overview on the flavour of programming in paclib. A detailed description of the user interface can be found in [5] the internal structure and the implementation of the system are described in [6] We assume that the reader is familiar with the concepts of saclib; ....

Peter A. Buhr and Richard A. Stroobosscher. The ¯System: Providing Lightweight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. Software --- Practice and Experience, 20(9):929--964, September 1990.

....al. 1992 ] is a large library of algebraic algorithms written in the C language. These algorithms are based on a list processing kernel that provides automatic garbage collection. Furthermore, the package supports exact arithmetic with arbitrary precision integers and rationals. 2. The uSystem [ Buhr and Stroobosscher, 1990 ] is a package that provides lightweight concurrency on mono processor workstations and shared memory multiprocessors running under the UNIX operating system. The System provides an efficient management of tasks i.e. light weight parallel processes. paclib combines the list processing and ....

....we believe that the main features have already been stabilized. The system has been implemented on a Sequent Symmetry shared memory computer with 20 processors running under the Dynix ptx version of UNIX System V.3. However, it is portable to any other machine supported by the System package [ Buhr and Stroobosscher, 1990 ] This report is intended as a detailed description of the internal structure of the paclib kernel. It explains how the system has been designed , not how it can be used . The actual user interface of paclib is described in [ Hong et al. 1992 ] 2 SACLIB saclib is a library of algebraic ....

[Article contains additional citation context not shown here]

Peter A. Buhr and Richard A. Stroobosscher. REFERENCES 22 The ¯System: Providing Light-weight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. Software --- Practice and Experience, 20(9):929--964, September 1990.

....when the user starts the execution of a program or attaches an existing file of event data. The resulting displays can have significant differences, particularly in the semantic significance of the lines and symbols used. Figures 2 and 3 show examples of displays obtained in the OSF DCE and C [2] environments, respectively. It is not possible to describe the displays fully, but a description of a few features will illustrate the differences that can exist because of differences in target environments. In the OSF DCE display, all trace lines represent threads (in clients or servers) in ....

P. A. Buhr and R. A. Stroobosscher. The ¯System: Providing light-weight concurrency on shared-memory multiprocessor computers running UNIX. Software--- Practice and Experience, 20(9):929--963, September 1990.

....of the standard can circumvent operational and performance problems. Threads have been proposed as basic units for enabling concurrency in many systems. For example, on shared memory multiprocessors, Ariadne multiplexes user threads on top of multiple processes. The PRESTO [4] system and System [6] also use processes in this manner, though the implementations are less general and do not support distributed memory architectures. Examples of other systems providing threads or threads like interfaces with distributed memory support include IVY [20] Amber [9] Clouds [11] and distributed ....

....Filaments [15] In IVY, a network wide virtual address space, shared among loosely coupled processors, permits computations to access remote objects. The shared virtual 2 For examples of threads systems see FastThreads [3] C Threads [39] Solaris Threads [31] Filaments [13] System [6], Pthreads [27] memory allows threads and objects to migrate easily, since all processors have access to the same virtual address space. But in IVY, this requires maintaining coherence of copies of data resident on distinct processors. Such maintenance overheads and contention for multiple ....

P. A. Buhr and R. A. Stroobosscher. The ¯System: Providing light-weight concurrency on sharedmemory multiprocessor computers running unix. Software--Practice and Experience, 20(9):929-- 964, September 1990.

....drawback to this approach is that POSIX was designed as an application program interface, with features such as real time scheduling support that may add overhead for parallel systems. A lower level interface designed specifically as a compiler target would most likely result in better performance [1] and will be investigated in future research. To summarize, the mapping of computation to physical processors is determined by both the mapping of threads to contexts and the mapping of contexts to nodes. The relationship between nodes, contexts, and threads is illustrated in Fig. 1. Global ....

Peter Buhr and R. Stroobosscher. The ¯system: Providing light-weight concurrency on sharedmemory multiprocessor systems running Unix. Software Practice and Experience, pages 929-- 964, September 1990.

....the following requirements: upward compatibility to the Saclib kernel, high level parallel programming model, light weight concurrency, non determinism and speculative parallelism, communication by heap references, and parallel garbage collection. For this purpose, we adapted the System package [Buhr and Stroobosscher, 1990] that supports light weight concurrency on shared memory multiprocessors and workstations running under UNIX. The system has been implemented on a 20 processor Sequent Symmetry shared memory multi processor. The visualization environment Pacvis [Hong et al., 1994] allows to monitor the dynamic ....

....architecture. In the following subsections, we will describe how and to which extent we met these demands. A comprehensive description of the kernel design can be found in [Schreiner, 1992] 4.2. Overall Design The task management of the Paclib kernel is built on top of the System [Buhr and Stroobosscher, 1990]. This free software package consists of a library of C functions supporting light weight concurrency on shared memory multiprocessors and Unix workstations. The System kernel distributes tasks (light weight processes) among virtual processors which are (heavy weight) Unix processes scheduled by ....

P. A. Buhr and R. A. Stroobosscher. The ¯System: Providing Light-weight Concurrency on SharedMemory Multiprocessor Computers Running UNIX. Software --- Practice and Experience, 20(9):929--964, September 1990.

....above. The prototype was carefully designed to limit the code that was dependent on properties of the target environment being debugged. The prototype originally worked with Hermes [10] but several other target environments are now supported: Concert C [14] OSF DCE [7] SR [1] the System [2], and the debugger itself. The debugger is organised internally as multiple processes and has been instrumented so that it can be used on itself. The extensions that have been made to the debugger for real time support are, in principle, available in all these target environments. Since, ....

Peter A. Buhr and Richard A. Stroobosscher. The ¯System: Providing light-weight concurrency on sharedmemory multiprocessor computers running UNIX. Software---Practice and Experience, 20(9):929--963, September 1990.

....Computation . computation, polynomial factorization, real root calculation and algebraic number arithmetic. Several algebraic applications have been implemented in this library such as the Partial Cylindric Algebraic Decomposition Method [10] and the Grobner Bases Algorithm [4] System [7] is a library that supports concurrency on shared memory multiprocessors and Unix workstations. The System kernel distributes tasks (light weight processes) among virtual processors which are Unix processes scheduled by the operating system. On a multi processor, virtual processors are therefore ....

P. A. Buhr and R. A. Stroobosscher. The ¯System: Providing Light-weight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. Software --- Practice and Experience, 20(9):929--964, September 1990.

No context found.

P. A. Buhr and R. A. Stroobosscher, `The System: providing light-weight concurrency on sharedmemory multiprocessor computers running UNIX', Software---Practice and Experience, 20, 929--963. (1990).

....all of the inter language factors. To avoid this problem, all of the monitors described here (except the restricted automatic signal monitor) have been implemented in C [Kernighan and Ritchie 1988] using a concurrency kernel that supports light weight tasks on uniprocessors and multiprocessors [Buhr and Stroobosscher 1990] . The monitors are implemented by a preprocessor [Fortier 1989] that converts the monitor statements into appropriate declarations of semaphores and calls to P and V operations using these semaphores. The simulation using semaphores is faithful to directly implemented monitors in that the amount ....

BUHR, P. A., AND STROOBOSSCHER, R. A. 1990. The ¯System: Providing light-weight concurrency on sharedmemory multiprocessor computers running UNIX. Software---Practice and Experience 20, 9 (Sept.), 929--963.

....constructs, the translator may not be able to automatically detect all possible concurrent situations that could appear in a program or that it may be too expensive to analyze an entire program to detect them all. Virtually all concurrency libraries, including the ones build by this author [BS90, BDS 92] are built on an unsafe foundation that can result in unexpected failure both inside the library or in user programs. To ensure correct operation, most of these libraries rely on the fact that many compiler optimizations do not cross a routine call boundary. However, with the advent ....

Peter A. Buhr and Richard A. Stroobosscher. The ¯System: Providing Light-Weight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. Software---Practice and Experience, 20(9):929--963, September 1990.

No context found.

P. A. Buhr and R. A. Stroobosscher. The System: Providing light-weight concurrency on shared-memory multiprocessor computers running Unix. Software --- Practice and Experience, 20(9):929--963, Sept. 1991.

No context found.

P. A. Buhr and R. A. Stroobosscher. The ¯System: Providing light-weight concurrency on shared-memory multiprocessor computers running UNIX. Software---Practice and Experience, 20(9):929--964, September 1990.

No context found.

Peter A. Buhr and Richard A. Stroobosscher. The ¯System: Providing Light-weight Concurrency on Shared-Memory Multiprocessor Computers Running UNIX. REFERENCES 28 Software --- Practice and Experience, 20(9):929--964, September 1990.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC