Citation Context

... not well defined in a distributed system. A more precise notion is sequential consistency, whereby the memory appears to all processes as if they were executing on a single multiprogrammed processor =-=[7]. With seq-=-uential consistency, the notion of "the last value written" is precisely defined. The simplicity of this model may, however, exact a high price in terms of performance, and therefore much re...

Citation Context

...sors dedicated for the purpose of providing computing cycles. Alternatively, it may consist of a dynamically varying set of machines on which idle cycles are used to perform long-running computations =-=[14]-=-. In the latter case, the (hardware) cost is essentially zero, since many organizations already have extensive workstation networks in place. In terms of performance, improvements in processor speed a...

Citation Context

...ions using the TreadMarks distributed shared memory (DSM) system. DSM allows processes to assume a globally shared virtual memory even though they execute on nodes that do not physically share memory =-=[9]-=-. Figure 1 illustrates a DSM system consisting of N networked workstations, each with its own memory, connected by a network. The DSM software provides the abstraction of a globally shared memory, in ...

Citation Context

...ming models. 9.1 Alternative Programming Models Message Passing (PVM). Currently, message passing is the prevailing programming paradigm for distributed memory systems. Parallel Virtual Machine (PVM) =-=[12]-=- is a popular software message passing package. It allows a heterogeneous network of computers to appear as a single concurrent computational engine. TreadMarks is currently restricted to a homogeneou...

Citation Context

...ory consistency models, and we have designed protocols to combat the false sharing problem. These approaches are discussed next. 4 Lazy Release Consistency 4.1 Release Consistency Release consistency =-=[5]-=- is a relaxed memory consistency model. It does not guarantee consistency at all times; the propagation of updates to shared memory may be delayed for some amount of time. By doing so, a release consi...

Citation Context

...action does not result in a large amount of communication. Various techniques have been used in TreadMarks to meet this challenge, including lazy release consistency [6] and multiple writer protocols =-=[3]-=-. This paper first describes the application programming interface provided by TreadMarks (Section 2), Next, we discuss the implementation challenges. (Section 3) and the techniques used to meet these...

Citation Context

...between the two environments. Finally, DSM allows for seamless integration of shared-memory multiprocessor workstations in a network environment. The actual system described in this paper, TreadMarks =-=[7]-=-, runs at user-level on Unix workstations. No kernel modifications or special privileges are required, and standard Unix compilers and linkers are used. The challenge in implementing a DSM system is t...

Citation Context

... sure that the shared memory abstraction does not result in a large amount of communication. Various techniques have been used in TreadMarks to meet this challenge, including lazy release consistency =-=[6]-=- and multiple writer protocols [3]. This paper first describes the application programming interface provided by TreadMarks (Section 2), Next, we discuss the implementation challenges. (Section 3) and...

Citation Context

... has to divide the computation among different threads and use either synchronization or message passing to control the interactions among the concurrent threads. With implicit parallelism, as in HPF =-=[6]-=-, the user writes a single-threaded program, which is then parallelized by the compiler. In particular, HPF contains data distribution primitives, which may be used by the compiler to drive the parall...

Citation Context

...te protocol works better than an update protocol, because of the large amount of data resulting from the update protocol. Entry Consistency (Midway). Entry consistency is another relaxed memory model =-=[3]-=-. As in release consistency, consistency actions are taken in conjunction with synchronization operations. Unlike release consistency, however, entry consistency requires that each shared data object ...

Citation Context

...ive approach to shared memory is to implement it in hardware, using a snooping bus protocol for a small number of processors or using a directory-based protocol for larger number of processors (e.g., =-=[8]-=-). We share with this approach the programming model, but our implementation avoids expensive cache controller hardware. On the other hand, a hardware implementation can efficiently support applicatio...

Citation Context

...ng solely on commodity hardware and software, networks of workstations offer parallel processing at a relatively low cost. A network-of-workstations multiprocessor may be realized as a processor bank =-=[17]-=-, a number of processors dedicated for the purpose of providing computing cycles. Alternatively, it may consist of a dynamically varying set of machines on which idle cycles are used to perform long-r...

Citation Context

...n operations can be mapped into acquires and releases as well. A partial order, happened-before-1 , denoted hb1 ! can be defined on releases, acquires, and shared memory accesses in the following way =-=[1]-=-: ffl If a 1 and a 2 are ordinary shared memory accesses, releases, or acquires on the same processor, and a 1 occurs before a 2 in program order, then a 1 hb1 ! a 2 . ffl If a 1 is a release on proce...

Citation Context

...te data to provide the abstraction of shared memory. Most DSM systems choose to replicate data, because this approach gives the best performance for a wide range of application parameters of interest =-=[11]-=-. With replicated data, the provision of memory consistency is at the heart of a DSM system: the DSM software must control replication in a manner that provides the abstraction of a single shared memo...

Citation Context

...on the chromosome can be computed. ILINK [4] is a parallelized version of a widely used genetic linkage analysis program, which is part Mendel Recombination Figure 12 DNA Recombination of the LINKAGE =-=[10]-=- package. ILINK takes as input a family tree, called a pedigree, augmented with some genetic information about the members of the family. It computes a maximum-likelihood estimate of `, the recombinat...

Citation Context

...omplish the same goal. For a more detailed comparison in programmability and performance between TreadMarks and PVM we refer the reader to Lu's M.S. thesis, which includes nine different applications =-=[10]-=-. 22 901 4682 774 4085 9570 Processors 1 2 3 4 5 6 7 8 Speedup 1 2 3 4 5 6 7 Figure 14 Speedup Results for ILINK: Each line represents a different data set. The numbers at the bottom indicate the sequ...

Citation Context

...parallelism, such as TSP, ILINK, or MIP are not easily expressed in the HPF framework. Object-Oriented Parallel Computing (Orca). Rather than providing the programmer with a shared memory space, Orca =-=[16]-=- and other object-oriented parallel computing systems support a shared space of object, each of which is accessed by properly synchronized methods. Besides the advantages from a programming perspectiv...

Citation Context

...at recombination has occurred between the gene we are looking for and genes with known locations. From these probabilities an approximate location of the gene on the chromosome can be computed. ILINK =-=[4]-=- is a parallelized version of a widely used genetic linkage analysis program, which is part Mendel Recombination Figure 12 DNA Recombination of the LINKAGE [10] package. ILINK takes as input a family ...

Citation Context

... to implement release consistency. Roughly speaking, lazy release consistency enforces consistency at the time of an acquire, in contrast to the earlier implementation of release consistency in Munin =-=[3]-=-, sometimes referred to as eager release consistency, which enforced consistency at the time of a release. Figure 6 illustrates the intuitive argument behind lazy release consistency. Assume that x is...

Citation Context

... after the pre-processing step, which is not (yet) parallelized. In addition to the problems from the MIPLIB library, the code was also used to solve a previously unsolved multicommodity flow problem =-=[2]-=-. The problem took roughly 30 CPU days on an 8-processor IBM SP-1 and also exhibited near-linear speedup. misc07 mod011 p0548 p6000 set1al stein45 1 2 3 4 5 6 7 8 0 2 4 6 8 10 12 14 Figure 11 Results ...

Citation Context

...n hardware may lead to poor performance because of discrepancies between the page size of the machine and the granularity of sharing in the application. The system discussed in this paper, TreadMarks =-=[5]-=-, provides shared memory as a linear array of bytes. The memory model is a relaxed memory model, namely release consistency. The implementation uses the virtual memory hardware to detect accesses, but...

Citation Context

...ilities an approximate location of the gene on the chromosome can be computed. ILINK is a parallelized version of a widely used genetic linkage analysis program, which is part of the FASTLINK package =-=[4]-=-. ILINK takes as input a family tree, called a pedigree, augmented with some genetic information about the members of the family. It computes a maximum-likelihood estimate of `, the recombination prob...

Citation Context

...s not well de ned in a distributed system. A more precise notion is sequential consistency, whereby the memory appears to all processes as if they were executing on a single multiprogrammed processor =-=[7]-=-. With sequential consistency, the notion of \the last value written" is precisely de ned. The simplicity of this model may, however, exact a high price in terms of performance, and therefore much res...

Citation Context

...on operations can be mapped into acquires and releases as well. A partial order, happened-before-1 , denoted hb1 ! can be de ned on releases, acquires, and shared memory accesses in the following way =-=[1]-=-: If a1 and a2 are ordinary shared memory accesses, releases, or acquires on the same processor, hb1 and a1 occurs before a2 in program order, then a1 ! a2. If a1 is a release on processor p1, and a2 ...

Citation Context

... to implement release consistency. Roughly speaking, lazy release consistency enforces consistency at the time of an acquire, in contrast to the earlier implementation of release consistency in Munin =-=[4]-=-, sometimes referred to as eager release consistency, which enforced consistency at the time of a release. Figure 7 shows the intuitive argument behind lazy release consistency. Assume that x is repli...

Citation Context

... to implement release consistency. Roughly speaking, lazy release consistency enforces consistency at the time of an acquire, in contrast to the earlier implementation of release consistency in Munin =-=[4]-=-, sometimes referred to as eager release consistency, which enforced consistency at the time of a release. Figure 7 shows the intuitive argument behind lazy release consistency. Assume that x is repli...

Citation Context

... to implement release consistency. Roughly speaking, lazy release consistency enforces consistency at the time of an acquire, in contrast to the earlier implementation of release consistency in Munin =-=[4]-=-, sometimes referred to as eager release consistency, which enforced consistency at the time of a release. Figure 7 shows the intuitive argument behind lazy release consistency. Assume that x is repli...