Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of the Sun User Group Technical Conference, pages 101--112, June 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....calculations can proceed while the processor s main thread performs the communication steps. The facilities offered by MD for synchronization, latency hiding in communications, and for creation and deletion of execution threads rely on the Cthreads parallel programming platform described elsewhere[Muk91] 2.3 Summary of Core MD The sections above describe the computational core of MD, which would serve as a starting point for abstracting a computational kernel from MD. However, as with many complex applications, this 5 computationally intensive core comprises less than one third of the ....

....by the KSR s OSF Unix operating system is one of kernel level threads which offer constructs for thread fork, thread synchronization and shared memory between threads. This kernel level thread facility is called Pthreads. MD itself is implemented with the Cthreads parallel programming library[Muk91] a user level threads facility that is built on top of Pthreads. As Cthreads performs its functions outside the kernel (within a Pthread) it has performance advantages over Pthreads concerning thread creation, context switches, and synchronization. User level thread implementations like ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of the Sun User Group Technical Conference, pages 101--112, June 1991.

....has been developed for verifying the approach and evaluating performance of the proposed parallel system. The current implementation of the simulator includes features of STM and interactions between processes. 7. 1 The Simulator The simulator uses a light weight threads package called CTHREAD[11]. A scheduler is used to dispatch scheduled events to processors. Scheduling is based on the smallest timestamp policy, i.e. the globally smallest timestamped event would execute first. Once an event is assigned to a processor, it does not release the processor unless it finishes its execution or ....

Bodhisattwa Mukherjee. A Portable and Reconfigurable Threads Package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991.

....in layer computations. An additional optimization used in our implementation is to pre create helper threads, pre map them to the appropriate processors, then simply wake up such threads when they are needed, using the low level conditional wait and signal primitives offered by the Cthreads library[22, 23]. Such wakeups are performed whenever layer processors have reached the point in time when terms A, B, and C have to be computed. Layer processors compute the term D themselves, while waiting for the completion of helper threads computing A, B, and C. The time integration of the transport equation ....

....to any shared memory parallel machine, currently including SUN Sparcstations, SGI uni and multiprocessors, and the KSR 1 and KSR 2 supercomputers. This portability is achieved by use of the Cthreads portable parallel programmming library described in detail in several publications, including [22, 23]. This library hides underlying differences in parallel machines and operating systems by providing a standard set of constructs for creating and controlling parallel execution threads. For example, for thread synchronization, mutex locks offered by the library are layered directly on low level ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991.

....implementation we chose as a prototype was Pthreads[Ken92] a standard thread library on KSR parallel machines. This platform was available to us locally and provided a good example of a threads implementation. Other similar threads implementations include Solaris threads from Sun and cthreads[Muk91]. We used POLKA[SK93] an object oriented software visualization methodology and library, as the platform to implement the graphical views. 2.1 Parallel Execution Environment KSR Pthreads and Gthreads The KSR Pthreads interface is based on draft 4 of the proposed POSIX Threads Extension for ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991.

....) to delimit the utility of Falcon in terms of its offered functionality and its associated performance characteristics, and (3) to demonstrate that the implementation of Falcon may be ported to a wide variety of target platforms. Falcon s implementation relies on a Mach compatible Cthreads library[38] available on a variety of hardware platforms, including the Kendall Square Research KSR 1 and KSR 2 supercomputers, the Sequent multiprocessor, uni and multi processor SGI, SUN SPARC, IBM RS6000 workstations, and various Linux machines. Falcon s implementation structure is depicted in Figure 5. ....

....In addition, we briefly describe Falcon s steering mechanisms and its on line display system, which consume the information obtained from the on line monitoring system. Portability and limitations. The current implementation of Falcon relies on the availability of Georgia Tech Cthreads[38] on the desired target machine and on the availability of shared memory between application processes and local monitoring threads. Local monitoring threads are written as Cthreads programs and rely on Falcon s event buffering mechanisms for event transfer. These buffering mechanisms require ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of the Sun User Group Technical Conference, pages 101--112, June 1991.

....of other performance display tools, Falcon also interfaces to custom displays and to systems for the creation of high quality 3D visualizations of program output data, like the SGI Explorer tools. 3. 3 System Implementation Falcon s implementation relies on a Mach compatible Cthreads library[36] available on several hardware platforms, including the Kendall Square Research KSR 1 and KSR 2 supercomputer, the GP1000 BBN Butterfly multiprocessor, the Sequent multiprocessor, and uni and multi processor SGI and SUN SPARC workstations. Figure 5 depicts the system s implementation. It is ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991.

....over the Solaris threads library [16] a POSIX compliant package which provides kernel supported, preemptive threads which will operate with true concurrency on a multiprocessor. ffl We also ported the thread interface to use the Cthreads package developed at the Georgia Institute of Technology [14]. This thread system is implemented as a user space library without preemptive scheduling, but supports concurrent execution on multiprocessors. It is supported on a number of architectures, of which we utilized the versions running on IBM RS 6000 workstations and Silicon Graphics machines running ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991.

....a new iteration, occur first in the second thread from the bottom and later in threads of more distant domains. To produce displays of this type, the important constructs in the programming model must be instrumented. In this case, we have instrumented the Cthreads parallel programming library[Muk91] so that every operation that can affect the state of a thread produces a record in an event stream. In order to produce a reasonable display, these events must contain accurate timestamps and they should not excessively disturb normal execution of the program. The next section will discuss the ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991.

....strategies. A reconfigurable lock object permits the explicit alteration (statically and or dynamically) of its waiting and scheduling 1 behaviors. Experiments with reconfigurable locks with a workload generator on a 32 node BBN Butterfly multiprocessor (using a multiprocessor version of Cthreads[Muk91] as the basis) demonstrate that the application performance gains due to dynamic reconfiguration outweigh any performance penalties caused by the added runtime costs. Furthermore, several commonly held beliefs regarding multiprocessor synchronization are verified: ffl For improved performance, ....

....expand them until a tour is found. Once a tour is found for any subproblem, the search space may be pruned by deletion of some unnecessary branches and leaf nodes. When all leaf nodes have been expanded and pruned, the lowest tour is the minimum tour of the graph. We have used a Thread library[Muk91] to implement the TSP algorithm as a collection of asynchronous cooperating threads[CS93] The threads cooperate through two shared abstractions: 1) a shared work queue of subproblems which stores the search space tree, and (2) a shared value representing the current best tour found so far. To ....

[Article contains additional citation context not shown here]

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991. Techreport no: GIT-ICS91 /02.

....by selection of specific edges from the graph and creation of children of the root. Once a tour is found for a subproblem, the search space may be pruned by deletion of the unnecessary branches and leaf nodes. The algorithm is implemented as a collection of asynchronous cooperating threads [Muk91] synchronizing using four locks [Muk94] most of which address the shared queue in which subproblems are stored. The performance of these three parallel applications, Cholesky, Integer Sort, and TSP (the synchronization pattern for the locks of these applications appear in [MS94, Muk94] using ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of the Sun User Group Technical Conference, pages 101--112, June 1991.

.... locks 2 used for the synchronization of multiple processes on the BBN Butterfly NUMA multiprocessor[MS93a, MS93b] Using artificial workloads[MS93a] and a representative multiprocessor application a parallel branch and bound program on a 32 node BBN Butterfly multiprocessor[Muk91], we have demonstrated that: ffl Off line configuration is important for use of alternative thread schedulers, such as for realtime vs. non real time applications[SZG91, GFS93a] Similarly, multiple thread schedulers may be employed concurrently for performing the scheduling of application threads ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991. TR# GIT-ICS-91/02.

....lock implementations and compares them with the costs of the operations provided by the reconfigurable lock object used in our research. The following measurements are taken on a 32 node BBN Butterfly GP1000 NUMA multiprocessor using a multiprocessor version of Cthreads as the basis[Muk91, SFG 91] Table 2 lists the latencies of the lock operations for different lock implementations available on the BBN multiprocessor (provided by the hardware, operating system and the Cthreads library) A local lock refers to a lock which is located in the local physical memory whereas a ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991.

....among object fragments, thereby exploiting their application level knowledge about the semantics of object invocations and the communication patterns among object fragments. DSA support is implemented as a runtime library layered on top of a Mach compatible Cthreads package developed by our group[7, 33]. As a result, a parallel program written with the DSA library consists of a set of independent threads interacting via DSA objects. Implementation of a DSA object itself involves (1) the storage of object fragments (state and code) at participating processors and (2) communication between these ....

....on two different target machines: 1) a 32 node GP1000 BBN Butterfly multiprocessor and (2) a 32 node Kendall Square Research KSR 1 supercomputer. The library s portability to those machines is due to its implementation on top of a portable lightweight threads package developed by our group[33]. 3 The DSA Library: Implementing Distributed Objects The DSA library facilitates the implementation of shared, distributed abstractions in parallel programs in two ways: 1. at user level, by providing application programs with a uniform interface to shared abstractions, and 2. at ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991.

....are managed by runtime library routines linked into each application. A thread management operation does not require an expensive kernel call. Furthermore, lightweight threads enable an application program to use a thread management system, most appropriate to the problem domain. Mach Cthreads [60, 173, 212], the University of Washington threads [165, 9] SunOS LWP and threads [113, 127, 189] are a few popular lightweight thread implementations. A lightweight thread generally executes in the context of a middleweight or a heavyweight thread. Specifically, the threads library schedules lightweight ....

....ffl A capability based interprocess communication facility. ffl Language support for RPC [76, 122, 123] ffl Support for remote file accesses between autonomous systems. 10 Mach is binary compatible with Berkley s Unix 4. 3 bsd release ffl Lightweight user level threads known as Mach Cthreads [60, 173]. ffl Miscellaneous other support like debuggers for multithreaded applications [51] exception handling [33] etc. Structurally, Mach is organized horizontally (developed using micro kernel technology) The Mach kernel is a minimal, extensible kernel which provides a small set of primitive ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of the Sun User Group Technical Conference, pages 101--112, June 1991.

....application program to use a thread management system which is most appropriate to the problem domain. Recently, thread libraries have become a common element of new languages and operating systems for shared memory multiprocessor and uniprocessors to support concurrent programming. Mach Cthreads[6, 16], the University of Washington threads[15, 2] POSIX Pthreads[7] SunOS LWP and threads[12, 13, 20] are a few popular lightweight thread implementations. We have implemented a portable user level thread package 1 [16] on several multiprocessor platforms such as BBN Butterfly parallel processor, ....

....and uniprocessors to support concurrent programming. Mach Cthreads[6, 16] the University of Washington threads[15, 2] POSIX Pthreads[7] SunOS LWP and threads[12, 13, 20] are a few popular lightweight thread implementations. We have implemented a portable user level thread package 1 [16] on several multiprocessor platforms such as BBN Butterfly parallel processor, Kendall Square Research parallel processor, Sequent Symmetry multiprocessor, Silicon Graphics multiprocessor and on several uniprocessors such as Sun 3 and Sun 4 families. Though these machines run implementations of ....

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proceedings of Sun User Group Technical Conference, pages 101--112, June 1991. TR# GIT-ICS-91/02.

No context found.

Bodhisattwa Mukherjee. A portable and reconfigurable threads package. In Proc. of the Sun User Group Technical Conference, pages 101--112, June 1991.

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