A. Chien, J. Dolby, B. Ganguly, V. Karamcheti, and X. Zhang. Supporting high level programming with high performance: The Illinois concert system. In Second International Workshop on High-level Parallel Programming Models and Supportive Environments, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Figure 8.1. Our approach is to ensure that accesses are only to the local address space level and that the effect of a local operation is reflected to a remote data item related to a local data. Our approach is different in many respects from the approaches that implement a global address space [9, 24, 34, 35, 48, 65, 101]. First, we distribute the data and keep a small amount of information that enables the retrieval of any data item in any address space at run time. Moreover, instead of offering a global memory view as with a virtual shared memory, we offer a local view of one address space only. Second, we use a ....

....for packing and unpacking data into mobile objects and for explicitly specifying message exchange. With our approach we isolate the user from these aspects. Finally, there are many object models that add concurrency constructs to the C programming language to support data parallelism [24,34,48]. These models support HPF like concurrency in an object oriented setting. In contrast with the fine grain HPF concurrency style, we address coarse grain concurrency. This chapter has presented a distributed object model for data parallel irregular applications. The model is based on the notion ....

[Article contains additional citation context not shown here]

A. Chien, J. Dolby, B. Ganguly, V. Karamecheti, and X. Zhang. Supporting High Level Programming with High Performance: The Illinois Concert System. In Proceedings of the Second International Workshop on High-Level Programming Models and Supportive Environments (HIPS '97), April 1997.

....back end is coupled to any number of language specific front ends; among the source languages supported are Cecil, C , Java, and Modula 3. scope statement statement throw statement procedure definition throwing procedure definition catch statement try statement 6 The Illinois Concert system [CDG97] also supports the representation of high level object oriented information, and extends these to also manage the additional complexity that comes with concurrency and parallel programming. 7. Summary The OSUIF libraries give the SUIF user the ability to express fundamental features of ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting High Level Programming with High Performance: The Illinois Concert System. In HIPS '97 Conference Proceedings --- High-level Parallel Programming Models and Supportive Environments (Workshop at IPPS '97), April, 1997.

....other hand, if a mes saging layer s guarantees are too strong (i.e. they provide more functionality than is generally needed) the messaging layer s common case performance may be needlessly degraded. Analysis of the literature and our ongoing studies to support fine grained parallel com puting [5, 12, 13, 14] have led to the conclusion that a low level messaging layer should provide the following key guarantees: Reliable delivery, In order delivery, and Control over scheduling of communication work (decoupling) As mentioned in the previous section, studies of communication software costs [12] ....

A. Chien, J. Dolby, B. Ganguly, V. Karamcheti, and X. Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, pages 15-24, April 1997.

....a key strength of Concert is its flexiblility in how it interfaces to the machine architecture. The Concert system was designed to tackle large programs and to be run on large scale architectures. It has already demonstrated good performance on distributed memory machines on a number of codes [7], 19] In this work, we are taking a new direction for the system in two ways. First, we have ported the Concert System to a large scaled shared memory system. Second, we have used the system to compile and run a Structured Adaptive Mesh Refinement (SAMR) code which is currently being used in ....

....at a high level. In a later section, we present descriptions of the primitives in more detail as well as some performance measurements. 2.3.1 System Design The following is a brief description of the Illinois Concert System implementation. More indepth descriptions are available at [15] and [7]. Keep in mind that these concepts are at the runtime system level, and are not exposed to the user of the system. The Illinois Concert System uses a dynamic multi threading execution model. The following terms are defined for purposes of our description. ffl Address Space An address space is a ....

[Article contains additional citation context not shown here]

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, April 1997.

.... item = new Item; item Item(id, d, v) ItemList l = new ItemList; l ItemList(item, inventory) inventory = l; Figure 1: Inventory Program (before Object Inlining) 3 Concert Compiler Our object inlining analysis and optimization implementation is done using the Illinois Concert System [10], a vehicle for research into advanced implementations for objectoriented languages. It implements the language ICC [11] a language with C like syntax and Java like semantics, which we use for our benchmarks. It also provides an aggressive analysis framework [35, 34] with two novel features: ....

A. Chien, J. Dolby, B. Ganguly, V. Karamcheti, and X. Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, pages 15-24, April 1997.

....precision. Therefore, the parallelism across meshes is dynamic, and is not analyzable during compilation or even initialization of the program. SAMR s dynamic parallelism must be exploited during the run of the program, making parallelization of the method a challenge. The Illinois Concert System [3] and ICC language [13, 12] together form a parallel programming environment geared towards tackling dynamically parallel codes. This system provides a variety of support for dynamic parallelism and distributed parallel data structures a global namespace, efficient finegrained threads, ....

....and operated upon either explicitly by the programmer or automatically by a compiler or runtime system. The Illinois Concert System is a programming environment that harnesses the benefits of COOP, with a goal of high performance. It consists of the ICC language [12] Concert compiler [4, 15, 3] and the Concert runtime system [9] Concert supports fine grained, concurrent object oriented programming on Actors [1] Computation is expressed as method invocations on objects or collections of objects. Concurrent method invocations operate against state stored in dynamically created thread ....

[Article contains additional citation context not shown here]

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, pages 15--24, April 1997.

....the other hand, if a messaging layer s guarantees are too strong (i.e. they provide more functionality than is generally needed) the messaging layer s commoncase performance may be needlessly degraded. Analysis of the literature and our ongoing studies to support fine grained parallel computing [5,14 16] have led to the conclusion that a low level messaging layer should provide the following key guarantees: M. Lauria, S. Pakin, A. Chien Efficient Layering: MPI over FM 7 0 10 20 30 40 50 60 16 32 64 128 256 512 Msg Size Bandwidth (MB s) Link Mgmt I O Bus Mgmt Flow Control (a) 0 ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, pages 15--24, April 1997.

....system, we have demonstrated the above techniques on numerous application programs and kernels. These demonstrations have repeatedly confirmed the benefits of high level programming constructs and aggressive implementation techniques for achieving high performance, flexible application software [8]. Application demonstrations are used to illustrate the good speedups and high absolute performance levels achieved, and also to assess the impact of various optimizations on program performance. 1.1 Organization The rest of this paper is structured as below. Section 2 describes the ICC ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, April 1997.

....proved insufficent. Fortunately, the adaptive interprocedural analysis we employed [24] computes precise information efficiently. We begin by discussing our approach to inline allocation in Section 2 and presenting an example program in Section 3. Next, Section 4 describes the Concert System [7] in which work was done, and Section 5 details our three program analyses. These analyses are evaluated on a suite of C programs in Section 6. This suite is summarized in Section 6.2 and performance metrics and results are given in Sections 6.3 and 6.4. Finally, we contrast related work in ....

....one from a free standing Point. These two x pos operations must have different implementations, as one must access the inlined field and the other a freestanding one. 4 Background: The Concert Compiler The implementation of our object inlining analyses was done in the Illinois Concert System [7], and so a brief discussion of the relevant aspects of the system is given here to provide context for subsequent description of the optimization. Most relevant is the program representation and the analysis and cloning frameworks, all discussed below. 4.1 Program Representation The primary ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, pages 15--24, April 1997.

....and non binding concurrency. These features support convenient expression of both regular and irregular concurrency, using an object oriented approach. In addition, we have developed a range of aggressive compiler and runtime implementation techniques embodied in the Illinois Concert system [4], which support the efficient execution of such irregular applications on a wide range of parallel architectures. Our previous published studies have focused on demonstrating high performance for individual applications. In this paper, we use our suite of applications and mature Concert system ....

....locality analysis to enable cheap access to local objects. Implicit concurrency control requires the system to ensure consistency with minimal overhead. Non binding concurrency forces the system to address issues of granularity and parallel overhead. We developed various optimization techniques [4] to address these issues. Specific techniques include: ffl Aggressive flow sensitive interprocedural analysis [17, 18] ffl Directed cloning and optimization (procedure and object inlining) 6, 20] ffl Compiler managed locality and memory latency management [27] ffl Efficient, robust ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, April 1997.

.... This result is encouraging because it is practical to obtain coarse grained aliasing information, and block level concurrency annotations are both simple to add to programs and already supported by various programming languages [5, 8, 16] We have implemented DPA in the Illinois Concert system 1 [7] in the context of optimizing concurrent object oriented programs relying on global, distributed PBDSs on parallel machines. We have experimentally evaluated our technique on the CRAY T3D, on the force computation phases of Barnes Hut and FMM, two hierarchical N body programs from the SPLASH 2 ....

....object.method 1( objects[i] method( object.method 2( The conc specification asserts the absence of indirect dependency among loop iterations or statements within a block and enables interleaved execution. 3. 3 Project Context We implemented DPA in the Illinois Concert compiler [7], which uses program dependence graph (PDG) 15] in gated SSA [12] form as its intermediate representation. The Concert compiler supports an adaptive, context sensitive interprocedural analysis framework [27, 28] that we use to implement our analysis and code transformations. The Concert compiler ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, April 1997.

....This requires dataflow analysis that can flow properties through fields (Sec Point: area(p) return abs(xpos p.xpos) abs(ypos p.ypos) Point: abs( return sqrt(xpos xpos ypos ypos) Rectangle: area( return lowerleft.area(upperright) Figure 4: Methods tion 3.2. 1) Our Concert [8] analysis framework is, to our knowledge, the only one that can do this. dorectangle(ll, ur) r = new Rectangle(ll, ur) cout r.area( ll = new List(r.lowerleft, nil) l2 = new List(r.upperright, nil) cout head(l1) abs( cout head(l2) abs( main( p1 = new Point(1.0,2.0) p2 = ....

....burden of deciding what to inline on the programmer. On the other hand, because we want to preserve a uniform model and our analysis opens up other optimization opportunities, our inline allocation is done automatically by the compiler. 3. 2 The Concert Compiler This works was done in the Concert [8] compiler, so subsequent discussion of our optimization relies heavily on the framework of that compiler. Therefore, a brief overview is given of the two portions that we use: the analysis and cloning modules. 3.2.1 The Analysis Framework The Concert compiler has a global analysis framework (see ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The illinois concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, April 1997.

....the other hand, if a messaging layer s guarantees are too strong (i.e. they provide more functionality than is generally needed) the messaging layer s commoncase performance may be needlessly degraded. Analysis of the literature and our ongoing studies to support fine grained parallel computing [5,14 16] have led to the conclusion that a low level messaging layer should provide the following key guarantees: ffl Reliable delivery, ffl In order delivery, and ffl Control over scheduling of communication work (decoupling) As mentioned in the previous section, studies of communication software ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, pages 15--24, April 1997.

....range of irregular applications. Our application study is done in the context of the Illinois Concert system, a high performance compiler and runtime for parallel computers which has been the vehicle for extensive research on compiler optimization and runtime techniques over the past five years [3]. Using the mature Concert system, we evaluate programming effort for a rich suite of irregular parallel applications. Our results indicate that a high level concurrent object oriented programming model and a sophisticated implementation can eliminate many concerns, easing the programming of ....

.... by a global priority scheduler: thread function( arg1, priority arg1 scheduler PRIORITY ) Implementation Technology The Concert system contains a wide range of aggressive optimizations and has been used to demonstrate high performance in absolute terms on a wide range of applications [3]. It automatically addresses several concerns which programmers must manage in lower level programming models: Procedure Granularity and Virtual Function Call Overhead Compound Object Structuring Thread Granularity Distributed Name Management Initial data placement, Data ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The Illinois Concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, April 1997.

....the other hand, if a messaging layer s guarantees are too strong (i.e. they provide more functionality than is generally needed) the messaging layer s common case performance may be needlessly degraded. Analysis of the literature and our ongoing studies to support fine grained parallel computing [12, 28, 29, 30] have led to the conclusion that a low level messaging layer should provide the following key guarantees: ffl Reliable delivery, ffl Ordered delivery, and ffl Control over scheduling of communication work (decoupling) Previous studies of communication cost in the CM 5 multicomputer system [28] ....

Andrew Chien, Julian Dolby, Bishwaroop Ganguly, Vijay Karamcheti, and Xingbin Zhang. Supporting high level programming with high performance: The illinois concert system. In Proceedings of the Second International Workshop on High-level Parallel Programming Models and Supportive Environments, April 1997.

No context found.

A. Chien, J. Dolby, B. Ganguly, V. Karamcheti, and X. Zhang. Supporting high level programming with high performance: The Illinois concert system. In Second International Workshop on High-level Parallel Programming Models and Supportive Environments, 1997.

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