P. E. Crandall, R. A. Aydt, A. A. Chien, and D. A. Reed, \Input/Output Characteristics of Scalable Parallel Applications," in Proceedings of Supercomputing '95, (San Diego, CA), pp. CD{ROM, IEEE Computer Society Press, Dec. 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....is NP complete. Input output characterization. There are a number of studies that present models of physical disk [23, 24, 25] and disk array [17] access behavior. Also, the logical and physical patterns of application input output in parallel scienti c applications have been studied extensively [9, 26, 8, 27, 10, 28]. These studies have shown that parallel applications exhibit a wide variety of input output request patterns. Insights from these studies led to new, parallel le system application programming interface (API) standardization e orts like the SIO API [29] and the MPI IO API [30] Flexible ....

P. E. Crandall, R. A. Aydt, A. A. Chien, and D. A. Reed, \Input/Output Characteristics of Scalable Parallel Applications," in Proceedings of Supercomputing '95, (San Diego, CA), pp. CD{ROM, IEEE Computer Society Press, Dec. 1995.

....A recent effort is the the development of visualization corridors [27] focusing an integrated solution for distance visualization. Turning to the I O issue, work has been done mainly addressing the I O characteristics of graphics and visualization applications on parallel computers [4, 23, 3]. The parallel I O capabilities of MPI 2 [6] offer a standard way to implement high performance I O which allows noncontiguous access patterns and collective I O. Kotz maintains an online archive (http: www.cs.dartmouth.edu pario ) providing abundant information about the research on and the use ....

P. E. Crandall, A. A. Ruth, A. A. Chien, and D. A. Reed. Input/output characteristics of scalable parallel applications. In Proceedings of Supercomputing '95, November 1995.

....Brown et al. 3] build a meta data system on top of HPSS using DB2 from IBM. The SRB [2] and MCAT [12] provide a uniform interface for connecting to heterogeneous data resources over a network. Three I Ointensive applications from the Scalable I O Initiative Application Suite are studied in [8]. But all these works only address one aspect of the issues we discussed above. Our previous work [6] is a first step toward considering multiple factors in an overall picture. We have designed an active meta data management system that take advantage of stateof the art I O optimizations as well ....

P. E. Crandall, R. A. Aydt, A. A. Chien, and D. A. Reed. Input/output characteristics of scalable parallel applications. In Proceedings of Supercomputing '95.

....le. This node reads the rst 2 MB of the le a single byte at a time and in sequential order. This application request pattern represents two interesting extremes. All accesses are purely sequential and are of the smallest possible size. 4.1. 4 Hartree Fock The Hartree Fock code (described in [3]) calculates non relativistic interactions between atomic nuclei and electrons in reaction paths that are not feasible to explore experimentally. Three distinct programs comprise this code and operate as a logical pipeline through which data ows. The third phase of the code has the greatest I O ....

....programs comprise this code and operate as a logical pipeline through which data ows. The third phase of the code has the greatest I O requirements, so we choose to focus our attention solely on this phase. The trace le we use was obtained from a run using 128 processors of an Intel Paragon [3]. Each processor repeatedly reads the integral les produced in phase two. We model the requests to one of these les from processor two. The initial o set for each data access is shown in Figure 4.6. Each request is for 80 KB of data, and the le is accessed sequentially a total of six times. In ....

Phyllis E. Crandall, Ruth A. Aydt, Andrew A. Chien, and Daniel A. Reed. Input/Output Characteristics of Scalable Parallel Applications. In Proceedings of Supercomputing '95, pages CD{ROM, San Diego, CA, December 1995. IEEE Computer Society Press.

....time of interactive tasks, and also improves its own performance through better replacement decisions and reduced memory management overhead. 1 Introduction Many of the computational problems of interest to scientists and engineers involve data sets that are much larger than physical memory [6, 7, 17]. Despite the continuing trend toward larger memories, it is unlikely that these data sets will ever fit entirely within main memory. Increases in processor power and memory capacity make it feasible to solve larger problems, or to solve the same problem at a finer granularity, but the size of the ....

P. E. Crandall, R. A. Aydt, A. A. Chien, and D. A. Reed. Input /Output Characteristics of Scalable Parallel Applications. In Proc. of Supercomputing '95, Dec. 1995.

....pace than I O technology. In fact, the strategy we develop in this research can be used in conjunction with parallel I O to achieve maximum performance. There has also been previous research investigating the I O characteristics of graphics and visualization 2 applications on parallel computers [5, 20]. Chiueh [3] presented a memory access algorithm that allows conflict free access to an interleaved memory system that stores volumetric data sets. The same algorithm is directly applicable in the context of parallel disk arrays. The work described here, in contrast, focuses mostly on resource ....

P. E. Crandall, A. A. Ruth, A. A. Chien, and D. A. Reed, Input/Output Characteristics of Scalable Parallel Applications, in Proceedings of Supercomputing '95, November 1995.

....involves performance related meta data as well which enables automatic high level I O optimizations as explained in this paper. There are several works on I O characterization of large applications. Three I O intensive applications from the Scalable I O Initiative Application Suite are studied in [12]. An I O intensive three dimensional parallel application code is used to evaluate the I O performances of the IBM SP and Intel Paragon in [28] They found IBM SP to be faster with read operations and Paragon for writes. del Rosario and Choudhary [15] discuss several grand challenge problems and ....

P. E. Crandall, R. A. Aydt, A. A. Chien, and D. A. Reed. Input/output characteristics of scalable parallel applications. In Proceedings of Supercomputing'95.

....in the common case. Unfortunately, large classes of applications exhibit file access patterns that are not amenable to sequential prefetching. In particular, applications from the database and scientific communities often perform non sequential, though regular, file access [17, 22] escat [6] is an example of an application utilizing strided access patterns and will be examined below in some detail. Such behavior is characteristic of data sets which are partitioned amongst processors so that the partitions may be processed independently of one another, as in stencil algorithms [11] ....

Phyllis E. Crandall, Ruth A. Aydt, Andrew A. Chien, and Daniel A. Reed. Input/output characteristics of scalable parallel applications. In Proceedings of Supercomputing '95, San Diego, CA, December 1995. IEEE Computer Society Press.

....also unique in considering persistent storage, complex distributions of data on processors and disks, and the use of a sophisticated run time system for optimizing resources. Several run time support libraries and file systems have been developed to support efficient I O in a parallel environment [7, 8, 13, 19]. They also usually provide a collective I O interface, in which all processing nodes cooperate to make a single large I O request. Our work is different in two important ways. First, we are supporting a much higher level of programming by involving a compiler. Second, our target run time system, ....

P. E. Crandall, R. A. Aydt, A. C. Chien, and D. A. Reed. Input /Output characteristics of Scalable Parallel Applications. In Proceedings Supercomputing '95, December 1995.

....to be issued to the file system. The file system can combine these small strided requests from different processors into a larger collective sequential request 5 . Cypher93] stated that at least one of the applications that was measured filled up main memory with data read sequentially. [Crandall95] presents two applications where, during the compulsory phase, entire files are read by one processing node. In the data staging phase, it is possible that the data is read with one access pattern and written with another. For example, the application can use a block column distribution of data ....

Phyllis E. Crandall, Ruth A. Aydt, Andrew A. Chien, and Daniel A. Reed. Input/output characteristics of scalable parallel applications. Proceedings of Supercomputing 1995 (San Diego, CA), pages 1643--73, December 1995.

....2 Related Out of Core Research 2. 1 Background A large body of literature relates to making use of parallel disks and improving the I O performance of large applications [39] A number of papers report the results of studies done to characterize disk access patterns of I O intensive applications [1, 24, 64, 59, 74]. The applications studied include many data parallel scientific and engineering applications, some database systems, data mining, and web servers, but not discrete event simulations. The results of these studies have motivated the design of algorithms, compilers, libraries, and file systems to ....

Phyllis E. Crandall, Ruth A. Aydt, Andrew A. Chien, and Daniel A. Reed. Input/output characteristics of scalable parallel applications. In Proceedings of Supercomputing '95, San Diego, CA, December 1995. IEEE Computer Society Press.

....facing scientists and engineers today involve solving problems with very large data sets. For example, global climate modeling, computational physics and chemistry, and many engineering problems (e.g. aircraft simulation) can easily involve data sets that are too large to fit in main memory [7, 9, 26]. For such applications (which are commonly referred to as out of core applications) main memory simply constitutes an intermediate stage in the memory hierarchy, and the bulk of the data must reside on disk or other secondary storage. Ideally one could efficiently solve an out of core problem ....

P. E. Crandall, R. A. Aydt, A. A. Chien, and D. A. Reed. Input/output characteristics of scalable parallel applications. In Proceedings of Supercomputing '95, December 1995. 72 BIBLIOGRAPHY 73

....Instead of optimizing the performance of the i o system, they took a different direction. They tuned the applications, e.g. reorganizing the order of the loops, to achieve high performance. Other works related to application level i o analyzed the i o characteristics of scientific applications [Crandall95, Pasquale94, Sio94, Thakur95] Understanding application i o requirements is important for providing any optimizations. Thakur95] studied the i o characteristics of an i o intensive parallel application in astrophysics and used the Chameleon library to perform i o. This application has the same ....

Phyllis E. Crandall, Ruth A. Aydt, Andrew A. Chien and Daniel A. Reed, Input/Output Characteristics of Scalable Parallel Applications, Proceedings of Supercomputing '95, December 1995

....from parallel computers. As a result, significant effort has been put into trying to improve parallel I O systems. To date, most of the effort has focused on observing the I O behavior of existing applications and on trying to improve the ability of I O systems to execute existing applications[2, 3, 4, 6, 8]. We decided to take a different approach; rather than tuning the I O system to optimize fixed applications, we concentrated on tuning the applications to improve their I O performance (and hopefully also improve their execution time) Our goal was to find out what strategies were required to ....

P. Crandall, R. Aydt, A. Chien, and D. Reed. Input/output characteristics of scalable parallel applications. In Proceedings Supercomputing'95, December 1995. To appear.

....analyzed for studying the file system s response [RBK92, KN95] 3. Application characterization: The third approach instruments the I O intensive applications and characterizes their access patterns. Most previous work has focussed on characterizing I O behavior of scientific applications [CACR95] Previous studies of DSM machines have addressed hardware and software design aspects [WOT 95, HSH96, BSF 91, SGA98] and their impact on scientific and commercial applications [KP98, ZS98] Related studies of the HP CONVEX machines include the architectural evaluations of the Convex ....

P. E. Crandall, R. A. Aydt, A. A. Chien, and D. A. Reed. Input/Output Characteristics of Scalable Parallel Applications. In Proceedings of Supercomputing '95, December 1995.

....These papers are detailed characterizations of the I O access pattern of one or more parallel applications. ffl [AUB 96] They discuss four application programs from the areas of satellite data processing and linear algebra. They tune each one of them by restructuring the program. ffl [CACR95] A detailed characterization of three applications: electron scattering, terrain rendering, and quantum chemistry. They look at the volume of data moved, the timing of I O, and the periodic nature of I O. They do a little bit with the access patterns of data within each file. They found a huge ....

Phyllis E. Crandall, Ruth A. Aydt, Andrew A. Chien, and Daniel A. Reed. Input /output characteristics of scalable parallel applications. In Proceedings of Supercomputing '95, December 1995.

....precision numbers per particle) Checkpointing. Checkpointing is often required for long running applications which can get interrupted for a variety of reasons [12] Checkpointing is also used for parametric studies, i.e. modifying some of the checkpointed values and restarting the computation [11]. Out of Core Computations. Several scientific and engineering computations operate on large data structures which do not fit into the main memory. Such codes need to access the secondary memory very frequently. Examples include several sensor data processing codes which perform out of core ....

....library for optimizing I O accesses and uses the two phase I O technique [7, 9] Kotz has developed a similar technique, disk directed I O for performing collective I O operations [18] Other projects have focussed on benchmarking I O intensive applications, including Crandall et. al at Illinois [11] and Acharya et. al at Maryland [1] An important limitation of our current work has been to consider each array as a single entity, i.e. modification or reference to any element of an array is considered as mod ref to the entire array. We plan to augment our analysis with array section analysis ....

P. E. Crandall, R. A. Aydt, A. C. Chien, and D. A. Reed. Input/Output characteristics of Scalable Parallel Applications. In Proceedings Supercomputing '95, December 1995. To appear.

....Scientific applications use larger files and have more sequential access [MK91, GGL93, PP93, PP94] Parallel scientific programs access the file with patterns not seen in uniprocessor or distributed system workloads. Although there seems to be a wide variety of access patterns [NKP 96, CACR95, SACR96] we have noticed many patterns accessing small, discontiguous pieces of the file in several types of strided pattern [NK96b, NKP 96] Finally, scientific applications use files for more than loading raw data and storing results; files are used as scratch space for large problems as ....

Phyllis E. Crandall, Ruth A. Aydt, Andrew A. Chien, and Daniel A. Reed. Input/output characteristics of scalable parallel applications. In Proceedings of Supercomputing '95, December 1995.

....and striping unit size, according to their impact on the I O performance of this application. 7 Related Work and Conclusions There are several works on I O characterization of large applications. Three I O intensive applications from the Scalable I O Initiative Application Suite are studied in [4]. An I O intensive three dimensional parallel application code is used to evaluate the I O performances of the IBM SP and Intel Paragon in [12] They found IBM SP to be faster with read operations and Paragon for writes. del Rosario and Choudhary [16] discuss several grand challenge problems and ....

P. E. Crandall, R. A. Aydt, A. A. Chien and D. A. Reed. Input/output characteristics of scalable parallel applications, In Proceedings of Supercomputing'95.

....fluid dynamics simulation. ffl PRISM is a three dimensional Navier Stokes turbulence simulation program written at Caltech [10] ffl RENDER is a program from JPL that combines satellite imagery with terrain elevation to produce three dimensional perspective views of Mars and Venus surfaces [16]. ffl PSTSWM is a benchmark program from Oak Ridge National Laboratory[25] that performs the core part of many weather modeling programs. All of the programs besides RENDER work in globally synchronized iterations. Thus it is straightforward to insert checkpoint here( calls at the end of ....

A. Chien P. Crandall, R. Aydt and D. Reed. Input/output characteristics of scalable parallel applications. In Proceedings of SuperComputing 95, 1995.

....via global pointers from the sensor actuator manager. By changing the mapping, one can apply the rule set using different sensors, choose different policies, or even control different systems. 8 Parallel Input Output Example Our recent characterization studies of parallel input output patterns [3,18,24,25,23,22] have shown that parallel applications exhibit a wide variety of input output request patterns, with both very small and very large request sizes, sequential and non sequential access, and a variety of temporal variations. Small input output requests are best managed by aggregation, prefetching, ....

Phyllis E. Crandall, Ruth A. Aydt, Andrew A. Chien, and Daniel A. Reed. Input/Output Characteristics of Scalable Parallel Applications. In Proceedings of Supercomputing '95, San Diego, CA, December 1995. IEEE Computer Society Press.

No context found.

P. E. Crandall, R. A. Aydt, A. A. Chien, D. A. Reed, Input/Output Characteristics of Scalable Parallel Applications, available at http://www-pablo.cs.uiuc.edu.

No context found.

Crandall, P., Aydt, R., Chien, A., Reed, D. Input/Output Characteristics of Scalable Parallel Applications. In Proceedings of Supercomputing 95, December 1995. To appear.

No context found.

P. E. Crandall, R. A. Aydt, A. A. Chien, D. A. Reed, Input/Output Characteristics of Scalable Parallel Applications, available at http://www-pablo.cs.uiuc.edu.

No context found.

P. E. Crandall, R. A. Aydt, A. A. Chien, D. A. Reed, Input/Output Characteristics of Scalable Parallel Applications, available at http://www-pablo.cs.uiuc.edu.

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