Katarzyna Keahey and Dennis Gannon. PARDIS: A parallel approach to CORBA. In HPDC, pages 31--39, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....CORBA applications or distributed client server or peer to peer parts in parallel MPI applications. Several independent projects have demonstrated CORBA s usefulness for parallel programming [12, 21] generally by extending the interactions available and supporting some level of data partitioning [15, 8]. This has typically resulted in non standard ORB extensions that are neither portable nor interoperable. But as the CORBA specification is not static, these project triggered the specification of Data Parallel CORBA [14] On the other end MPI programs are mostly efficient when it comes to data ....

....techniques are required but not sufficient for a strict zero copy implementation of an ORB. Integration of MPI in CORBA A first option to optimize bulk data handling in CORBA programs would be to integrate a full blown message passing interface library into an ORB. This route is taken by PARDIS [8]. This employs the key idea of CORBA interoperability through meta language interfaces to implement application level interaction of heterogeneous parallel components in a distributed environment. However, PARDIS extends the CORBA object model by introducing SPMD objects representing ....

K. Keaheya and D. Gannon. PARDIS: A Parallel Approach to CORBA. In In Proceedings of the 6th International Symposium of High Performance Distributed Computation (HPDC), pages 31--39. IEEE, Aug 1997.

....we have been about to link parallel HPC program running on two di erent architectures together as one single parallel program. In addition, we have been able to wrap Fortran MPI programs with HPC so that they can be accessed by HPC parallel programs running in di erent locations (see [52, 57]) Parallel Java and Interoperability with Remote Objects We have developed a Java preprocessor that will allow Java programs that are written with loops and certain types of recursion to be transformed into multi threaded code that executes in parallel on shared memory multiprocessors. As with ....

....we have focused many of our e orts on pushing the performance of CORBA in high performance, parallel application. We have developed a parallel CORBA ORB called PARDIS which allows SPMD object implementations to communicate with other similar application using parallel communication channels [57]. In particular, PARDIS extends the concept of a CORBA sequence to a distributed sequence. This system has been used to link parallel Fortran MPI applications with HPC SPMD computations. Recent Progress During the nal year of this project we have completed a second version of the Java RMI to ....

Kate Keahey and Dennis Gannon. PARDIS: A parallel approach to CORBA. In 6th IEEE International Symposium on High Performance Distributed Computation, August 1997.

....object, and using CORBA to manage the inter communication among objects. The advantage is that users can use existing CORBA implementations (such as Visibroker, Orbacus and others) without any modi cation to CORBA IDL compilers, as is done in other projects with a similar objective, such as PARDIS [10] and Cobra [9] Components in the CB PSE can be divided into four types: 1) compute components are responsible for computational work such as performing a matrix multiplication or a fast Fourier transform, 2) template components require one or more components to be inserted into them in order ....

K.Keahey and D. Gannon. PARDIS: A Parallel Approach to CORBA. In Proceedings of the 6th IEEE International Symposium of High Performance Distributed Computation, pp. 31-39, August 1997.

.... the efficiency of current distributed object s communication [6, 2] extending the real time and QoS features [11, 14] and supporting the mobile environment in a distributed object framework [13] Some frameworks support the composition of parallel component, such as LSA [3] POOMA [10] PARDIS [8], etc. DoHPC differs Multi threaded ORB in two aspects. Firstly, DoHPC provided partial concurrency control through inter method analysis, which makes the achieving of parallelism in a distributed object more transparent to its clients. Secondly, the implementation of multi threaded ORB is based ....

K. Keahey, Dennis Gannon, PARDIS: A Parallel Approach to CORBA, Technical report, ftp://ftp.cs.indiana.edu/pub/techreports/TR475.ps.Z, February 1997.

.... servers have also been implemented for several commercial ORBs [6, 7] An engineering solution for multithreading CORBA clients has been proposed by Hellemans et al. 8] Attempts at extending the CORBA object model to handle data parallel computations have been introduced by Keahey and Gannon [9]. Despite this growing interest, no attempt, has so far, been made to explicitly quantify the performance improvements obtained from different server implementation techniques. No optimizations have been investigated to combine the transparency characteristics of CORBA with the high performance ....

K. Keahey and D. Gannon. Pardis: A parallel approach to corba. In International Symposium on High Performance Distributed Computing, pages 31--9, Los Alamitos, CA, USA, Aug 1997. IEEE Comput. Soc.

....groups are currently working at projects to create distributed high performance environments. Some examples are COVISE [7]or CAESAR[8] which demonstrate the possibility of integrated distributed simulation systems. An approach to improve the data transfer rate in parallel applications is PARDIS [9], a system which uses CORBA for communication between different components and which extends the CORBA object model with Single Program Multiple Data (SPMD) objects representing data parallel computations. Further important approaches are GLOBUS [10] and the World Wide Virtual Machine (WWVM, 11] ....

Katarzyna Keahey and Dennis Gannon. Pardis: A parallel approach to corba. In Proceedings of IEEE 6th International Symposium on High Performance Distributed Computing, Portland, OR., August 1997.

....we have been about to link parallel HPC program running on two different architectures together as one single parallel program. In addition, we have been able to wrap Fortran MPI programs with HPC so that they can be accessed by HPC parallel programs running in different locations (see [52, 57]) Parallel Java and Interoperability with Remote Objects We have developed a Java preprocessor that will allow Java programs that are written with loops and certain types of recursion to be transformed into multi threaded code that executes in parallel on shared memory multiprocessors. As with ....

....we have focused many of our efforts on pushing the performance of CORBA in high performance, parallel application. We have developed a parallel CORBA ORB called PARDIS which allows SPMD object implementations to communicate with other similar application using parallel communication channels [57]. In particular, PARDIS extends the concept of a CORBA sequence to a distributed sequence. This system has been used to link parallel Fortran MPI applications with HPC SPMD computations. Recent Progress During the final year of this project we have completed a second version of the Java RMI to ....

Kate Keahey and Dennis Gannon. PARDIS: A parallel approach to CORBA. In 6th IEEE International Symposium on High Performance Distributed Computation, August 1997.

....raised in multiple contexts. Java based integration has been discussed by the Parallel Compiler Runtime Consortium [14] and implemented in the Converse framework [3] CORBA based integration asks for describing data distribution at the IDL level, and defining efficient communication mechanisms [12]. The PPADD (Parallel Predictive Adaptive Domain Decomposition) project is a large scale application that naturally requires a distributed implementation on heterogeneous parallel machines. The general application field is solving PDEs through adaptive domain decomposition methods. The entry point ....

K. Keahey and D. Gannon, PARDIS: A Parallel Approach to CORBA, Proceedings of the 6th IEEE International Symposium on High Performance Distributed Computing (best paper award), August 1997.

....a brief comparison with the Corba component model. 2.1 CCA The CCA [3] is an initiative by DOE laboratories and universities to develop a common architecture for scientific components. It aims to factor out the common functionality of current tools such as CUMULVS [28] PAWS [21] and PARDIS [16] into a standard platform for developing high performance components. Though originally intended to provide a source level platform, the CCA is now moving to defining a binary level platform. This will allow a compiled component to run on any conforming implementation. CCA components can be ....

Katarzyna Keahey and Dennis Gannon. PARDIS: A Parallel Approach to CORBA. In Proceedings of the 6th IEEE International Symposium on High Performance Distributed Computing, August 1997.

No context found.

K. Keahey and D. Gannon, "PARDIS: A Parallel Approach to CORBA," Proceedings of the 6 th IEEE International Symposium on High Performance Distributed Computation, August 1997.

....information subsystem, which also provides primitive archiving capabilities. Several research directions are extending the LSA. Among these are the development of methods for components implementing parallel algorithms for this a critical problem is parallel communications between components [14]. The LSA system currently has methods for gathering statistics and event logging for analysis and evaluation of the component system itself. 10 The availability of event logging allows the addition of reasoning tools, which can learn from earlier runs of the LSA to help guide a user for future ....

Kate Keahey and Dennis Gannon. PARDIS: A Parallel Approach to CORBA. In 6th IEEE International Symposium on High Performance Distributed Computation, August 1997.

....those offered by sequential languages lead to the development of parallel languages and libraries. From the point of view of parallel programming, the most important shortcoming of the existing component systems is that they don t support abstractions necessary for high performance programming [8] and they don t place enough emphasis on performance crucial in parallel programming [5] In the case of high performance systems, the latter problem is of particular importance; it is crucial that the multi component application developer has access to not only efficient components and efficient ....

....Component contracts describe which inputs need to be provided in order for the component to produce specific output or group of outputs. A component interface can contain several such contracts. Multi dimensional distributed grids and distribution specifications similar to those introduced in [8]. Both PARDIS and PAWS showed how the knowledge of distribution can be used to speed up data transfer between interacting SPMD components. Furthermore, 10] shows that knowing component distribution at the time of building an application (as opposed to finding out at run time) allows the ....

K. Keahey and D. Gannon. PARDIS: A Parallel Approach to CORBA. In Proceedings of the 6th IEEE International Symposium on High Performance Distributed Computation, pages 31--39, August 1997.

....discussion) Our approach leverages this work where appropriate, but addresses the distinctly different technical challenges of large scale scientific simulations. Based on the lessons learned from research projects in high performance component architectures by CCA participants (see, e.g. [3, 44, 25, 32, 36, 37]) and projects considering related design issues (see, e.g. 1, 23, 26, 6] we are developing a single component interface specification that will enable interactions among scientific components that follow this standard. Additional related work [10, 8, 22, 35] can be found elsewhere. We ....

....as HPC [24] POOMA [4] ISIS [12] SAMRAI [29] and PETSc [5] testifies that such abstractions can enable the user to develop more efficient programs faster. Similarly, we need abstractions that capture high performance concepts in component architectures. For example, PARDIS [37] and PAWS [6] successfully show that introducing abstractions for single program multiple data (SPMD) computation can enable more efficient interactions between SPMD programs. In this section, we briefly review these industry standards and explain their limitations for high performance scientific ....

[Article contains additional citation context not shown here]

K. Keahey and D. Gannon. PARDIS: A Parallel Approach to CORBA. In Proceedings of the 6th IEEE International Symposium on High Performance Distributed Computation, pages 31--39, August 1997.

....to recognize when a high speed pathway exists between two components (such as an ATM link) and adapt to the protocol supporting it. In particular, for components that are based on parallel implementations, the communications channels may need to be implemented as multiple concurrent pathways (see [11] for an example. A second problem with current commercial desktop component systems is the mechanism for encapsulating an existing scientific application (which may itself be a parallel distributed computation) into a component. Completely Java based systems are not suitable for an application ....

Kate Keahey and Dennis Gannon. PARDIS: A Parallel Approach to CORBA. In 6th IEEE International Symposium on High Performance Distributed Computation, August 1997.

....side (m) n=1 overhead send gather scatter Figure 5. Cost components of invocation using the centralized method with distributed arguments composed of doubles; N = 2 17 . The results for multi port method show that invocation times have an overall decreasing tendency with increasing n and m. In [KG97a] we have examined the different cost components of the invocation time and shown that in the presence of more than one active connection between client and server the sends are interleaved so that more than one communication operation can be active at the same time. This fact partially ....

K. Keahey and D. Gannon, PARDIS: A Parallel Approach to CORBA, Proceedings of the 6th IEEE International Symposium on High Performance Distributed Computation, August 1997.

....the information subsystem, which also provides primitive archiving capabilities. Several research directions are extending the LSA. Among these are the development of methods for components implementing parallel algorithms for this a critical problem is parallel communications between components [14]. The LSA system currently has methods for gathering statistics and event logging for analysis and evaluation of the component system itself. The availability of event logging allows the addition of reasoning tools, which can learn from earlier runs of the LSA to help guide a user for future ....

Kate Keahey and Dennis Gannon. PARDIS: A Parallel Approach to CORBA. In 6th IEEE International Symposium on High Performance Distributed Computation, August 1997.

....to recognize when a high speed pathway exists between two components (such as an ATM link) and adapt to the protocol supporting it. In particular, for components that are based on parallel implementations, the communications channels may need to be implemented as multiple concurrent pathways (see [13] for an example. A second problem with current commercial desktop component systems is the mechanism for encapsulating an existing scientific application (which may itself be a parallel distributed computation) into a component. Completely Java based systems are not suitable for an application ....

Kate Keahey and Dennis Gannon. PARDIS: A Parallel Approach to CORBA. In 6th IEEE International Symposium on High Performance Distributed Computation, August 1997.

....and tested components. This approach allows for a high level of component reusability and does not require existing components to be reimplemented. Further, it allows PARDIS to take advantage of application level information such as distribution of data structures in a data parallel program [KG97] PARDIS builds on CORBA in that it allows the programmer to construct metaapplications without concern for component location, heterogeneity of component resources, or data translation and marshaling in communication between them. However, PARDIS extends the CORBA object model by introducing ....

....the distribution of any of the in arguments to its operations prior to object registration; the client can set the distribution of the expected out arguments before making an invocation. Knowledge of distribution allows the ORB to efficiently transfer arguments between the client and server [KG97] The sequence class overloads operator[ to provide access to its elements with location transparency. We would like to stress however, that the main purpose of a distributed sequence is to be used as a container for argument data, not to provide its management. To that end, the distributed ....

K. Keahey and D. Gannon, PARDIS: A Parallel Approach to CORBA, Proceedings of the 6th IEEE International Symposium on High Performance Distributed Computation, August 1997.

No context found.

Katarzyna Keahey and Dennis Gannon. PARDIS: A parallel approach to CORBA. In HPDC, pages 31--39, 1997.

No context found.

Katarzyna Keahey and Dennis Gannon. PARDIS: A parallel approach to CORBA. In HPDC, pages 31--39, 1997. 12

No context found.

K. Keahey and D. Gannon. "PARDIS: A Parallel Approach to CORBA". In proc. of the 6th IEEE International Symposium on High Performance Distributed Computing, August 1997.

No context found.

K. Keahey, D. Gannon, PARDIS: A Parallel Approach to CORBA, in: Proceedings of the 6th IEEE International Symposium on High Performance Distributed Computation, 1997, pp. 31--39. 21

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