S. A. Moyer and V. S. Sunderam, "PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments," Proceedings of the Scalable High-Performance Computing Conference, pp 71-78, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....cluster. In order to effectively utilize this resource, applications must judiciously spread their I O requirements across the nodes in the cluster. In order to support these applications a number of researchers have begun to investigate distributed parallel file systems. Examples include PIOUS [2] and PVFS [8] These file systems are modeled after parallel file systems designed for parallel supercomputers such as IBM s Vesta [3] Intel s PFS, and Thinking Machine s CMMD I O [9] File systems for a DCPC differ from these primarily in that they are intended for a distributed computing ....

S. A. Moyer and V. S. Sunderam, "PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments," Proceedings of the Scalable High-Performance Computing Conference, pp 71-78, 1994.

....name representing a processor, a process, a location, etc. 2.3. Distributed I O 60 2. 3 Distributed I O This section describes two works incorporating I O capability for message passing environments, respectively MPI(Message Passing Interface) CFF 95] and PVM (Parallel Virtual Machine) MS94] The reason for describing these message passing proposals is to make clear that the problem dealt with in this thesis for LINDA has been identified and solved in other platforms. Once these proposals for message passing are described an outline of a more direct LINDA work is discussed [Gel89] ....

....I O systems on which MPI can rely. The MPI IO is mainly concerned with performance and therefore functionality is left as a second priority. This thesis takes the opposite view; it is concerned with functionality and also with unifying concepts. The PIOUS (Parallel Input OUtput System) MS94] is based on the inclusion of an I O capability within PVM. As with MPI IO, Moyer and Sunderam have identified the lack of a standard I O system for distributed environments like PVM. PIOUS is actually a parallel file system, taking the view that I O occurs via operations with files (like the ....

S. A. Moyer and V. S. Sunderam. PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments. In Proc. of the Conference on Scalable High-Performance Computing, pages 71--78. IEEE Computer Society Press, May 1994.

....with le systems designed for distributed access; particularly, distributed le systems are not designed for high bandwidth concurrent writes that parallel applications typically require. 2 A number of research projects exist in the areas of parallel I O and parallel le systems, such as PIOUS [19], PPFS [14, 26] and Galley [22] PIOUS focuses on viewing I O from the viewpoint of transactions [19] PPFS research focuses on adaptive caching and prefetching [14, 26] and Galley looks at disk access optimization and alternative le organizations [22] These le systems may be freely available ....

....for high bandwidth concurrent writes that parallel applications typically require. 2 A number of research projects exist in the areas of parallel I O and parallel le systems, such as PIOUS [19] PPFS [14, 26] and Galley [22] PIOUS focuses on viewing I O from the viewpoint of transactions [19], PPFS research focuses on adaptive caching and prefetching [14, 26] and Galley looks at disk access optimization and alternative le organizations [22] These le systems may be freely available but are mostly research prototypes, not intended for everyday use by others. 3 Design and ....

Steven A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable High-Performance Computing Conference, pages 71-78, 1994.

....highlighting that it would generalise the Linda model. Finally we conclude the paper in Section 7. 2. Related Work We rst describe two works incorporating I O capability for message passing environments, respectively MPI (Message Passing Interface) 4] and PVM (Parallel Virtual Machine) [10]. The reason for describing these message passing proposals is to make clear that the problem we intend to deal with in Linda has been identi ed and solved in other platforms. Once we describe these proposals for message passing, we outline a more direct Linda work [5] The basic idea of Corbett ....

....I O systems on which MPI can rely. The MPI IO is mainly concerned with performance and therefore functionality is left as a second priority. In our case we take the opposite view, we are concerned with functionality and also with unifying concepts. The PIOUS (Parallel Input OUtput System) [10] is based on the inclusion of an I O capability within PVM. As with MPI IO, Moyer and Sunderam have identi ed the lack of a standard I O system for distributed environments like PVM. PIOUS is actually a parallel le system, taking the view that I O occurs via operations with les (like the UNIX ....

S. A. Moyer and V. S. Sunderam. PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments. In Proceedings of the Conference on Scalable High-Performance Computing, pages 71-78. IEEE Computer Society Press, May 1994.

....several new parallel file systems based on user level libraries have emerged. We believe PPFS has a richer interface than these systems, supporting application control and extension of input output system policies, and declaration of access information. PIOUS (Parallel Input OUtput System) [20] is a portable input output system primarily for use with PVM. Unlike PPFS, PIOUS enforces sequential consistency on file accesses to provide fault tolerance on unreliable, distributed systems. In PPFS, because the primary focus is on parallel systems, data consistency can be controlled by the ....

Moyer, S. A., and Sundaram, V. S. PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments. In 1994 Scalable High Performance Computing Conference (May 1994), pp. 71--78.

....FDDI bandwidth of 12.5 MB sec using MPICH 1.0.11 [MPICH] 1 . As the number of senders and receivers increases, bandwidth per node degrades although the aggregate bandwidth increases. Thus interconnect speed, not disk speed, will often be the limiting factor for parallel I O performance. PIOUS [Moyer94], VIP FS [Harry95] and VIPIOS [Brezany96] provide parallel I O on workstation clusters. However, each of these lacks one or more of the features desired for parallel scientific applications running on a network of workstations: collective I O, special considerations for low message passing ....

....as its interface for write intensive applications on massively parallel platforms. Technology transfer from the Panda project has helped lead to an implementation of multidimensional array chunking, coupled with compression, in the HDF scientific data management package [Velamparampil97] PIOUS [Moyer94] is a pioneer work in parallel I O on a workstation cluster. PIOUS is a parallel file system with a Unix style file interface; coordinated access to a file is guaranteed using transactions. Our strategies for collective I O and for slow shared media networks could be built on top of PIOUS. VIP FS ....

S. Moyer and V. Sunderam, PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments, Computer Science Technical Report CSTR-940302, Department of Math and Computer Science, Emory University, November 1994.

....using different data 147 distributions. The optimal data distribution is obtained from finding the minimum of a performance prediction function. Data distribution on multi disk systems matches the two dimensional file structure supported by current parallel file systems, such as the PIOUS system [61] on network connected work stations and the PIOFS file system on the IBM SP 2 [20] Hence, the framework of program synthesis presented in this paper can be used to build a program development tool for implementing out of core applications on modern distributed hierarchical memory machines. In the ....

S. A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for 178 distributed computing environments. In Proceedings of the Scalable HighPerformance Computing Conference, pages 71--78, 1994.

....the portable parallel le systems (PPFS II) designed to support ecient access of large data sets in scienti c applications with irregular access patterns. More information on parallel and distributed I O and le systems appears in (Kotz ; Carretero et al. 1996; Gibson et al. 1999; Initiative ; Moyer and Sunderam 1994; Nieuwejaar and Kotz 1997; Schikuta et al. 1998; Seamons and Winslett 1996) Users of PKDD systems are interested in maximizing performance. Prefetching is an important performance enhancing technique that can reduce the impact of latency by overlapping computation and I O (Cortes 1999; Kimbrel ....

S. A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Scalable High-Performance Computing Conf., 1994.

....file systems designed for distributed access; particularly, distributed file systems are not designed for high bandwidth concurrent writes that parallel applications typically require. 2 A number of research projects exist in the areas of parallel I O and parallel file systems, such as PIOUS [19], PPFS [14, 26] and Galley [22] PIOUS focuses on viewing I O from the viewpoint of transactions [19] PPFS research focuses on adaptive caching and prefetching [14, 26] and Galley looks at disk access optimization and alternative file organizations [22] These file systems may be freely ....

....for high bandwidth concurrent writes that parallel applications typically require. 2 A number of research projects exist in the areas of parallel I O and parallel file systems, such as PIOUS [19] PPFS [14, 26] and Galley [22] PIOUS focuses on viewing I O from the viewpoint of transactions [19], PPFS research focuses on adaptive caching and prefetching [14, 26] and Galley looks at disk access optimization and alternative file organizations [22] These file systems may be freely available but are mostly research prototypes, not intended for everyday use by others. 3 Design and ....

Steven A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable High-Performance Computing Conference, pages 71--78, 1994.

....account for most of observed parallel I O patterns [24] In addition to the commercial offerings (IBM SP2 PIOFS [6] Intel iPSC CFS [25, 27] and Paragon PFS [11, 28] nCUBE [8] and Thinking Machines CM 5 sfs [2, 20] there has been a recent flurry of activity in the research community. PIOUS [22, 23] and PETSc Chameleon I O [14] are both widely available nonproprietary portable parallel I O interfaces. PIOUS is a PVM based parallel file interface. Files can be declustered across disks in a round robin fashion. Access modes support globally shared and independent file pointers, and file per ....

Steven A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable High-Performance Computing Conference, pages 71--78, 1994.

....available networks of workstations. A data storage and retrieval infrastructure needs to be constructed which will satify data access rates and capacities required by I O intensive programs. Only recently has any attempt been made at providing I O extensions to message passing libraries [12] [14]. Although these works recognized the deficiency in message passing libraries, they only constitute partial solutions. In order to deal with this issue in a general way, two problems need to be addressed: first, the problem of designing a parallel I O system with a coherent distributed, ....

S.A. Moyer and V.S. Sunderam. PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments. In Scalable High-Performance Computing Conference, May 1994.

.... one the same performance will be achieved Are there better placement algorithm to be used when a RAID is not homogeneous Which ones are them ffl How important is to place files in the disk which is closer to the application using it ffl When partitioning files (as done in Vesta [4] or PIOUS [5]) among heterogeneous disks, how should the sizes of the partitions be ffl How important is the network when implementing a distributed storage system 1.2 Related Work As far as we know, there are no tools as flexible as the one we present in this paper. All the simulators we have found are ....

....for clusters of workstations, we want to be able distribute data among disks emulating any system one can think of. We want to be able to simulate RAIDs, systems that place files in the disk that it is closer to the node that most frequently use them, or even file partitioning as done by PIOUS [5] or Vesta [4] This software hardware which is in charge of distributing data among disks is what we have called a distributor. A parallel distributed file system would be a software distributor while a RAID card controller would be a hardware distributor. Our simulator does not make any ....

S. A. Moyer and V. S. Sunderam. PIOUS: A scalable parallel I/O system for distributed computing environment. In Proceedings of the Scalable High Performance Computing Conference, pages 71--79, 1994.

....message passing libraries only provide part of the support necessary for most high performance distributed computing applications support for high speed parallel I O is still lacking. Only recently has any attempt been made at providing I O extensions to message passing libraries [11] [13]. Although these works recognized the deficiency in message passing libraries, they only constitute partial solutions. In order to deal with this issue in a general way, two problems need to be addressed: first, the problem of designing a parallel I O system with a coherent distributed, concurrent ....

S.A. Moyer and V.S. Sunderam. PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments. In Scalable High-Performance Computing Conference, May 1994.

.... one the same performance will be achieved Are there better placement algorithm to be used when a RAID is not homogeneous Which ones are them ffl How important is to place files in the disk which is closer to the application using it ffl When partitioning files (as done in Vesta [4] or PIOUS [5]) among heterogeneous disks, how should the sizes of the partitions be ffl How important is the network when implementing a distributed storage system 1.2 Related Work As far as we know, there are no tools as flexible as the one we present in this paper. All the simulators we have found are ....

....for clusters of workstations, we want to be able distribute data among disks emulating any system one can think of. We want to be able to simulate RAIDs, systems that place files in the disk that it is closer to the node that most frequently use them, or even file partitioning as done by PIOUS [5] or Vesta [4] This software hardware which is in charge of distributing data among disks is what we have called a distributor. A parallel distributed file system would be a software distributor while a RAID card controller would be a hardware distributor. Our simulator does not make any ....

S. A. Moyer and V. S. Sunderam. PIOUS: A scalable parallel I/O system for distributed computing environment. In Proceedings of the Scalable High Performance Computing Conference, pages 71--79, 1994.

....streams. Here we extend this work to include automatic mode detection, design details and tradeoffs, and optimizations. Most proposed parallel file systems for MIMD machines give the programmer one view of a file, namely as a single stream of bytes. Some systems, including Vesta [3] PIOUS [19] and others [12] support multifiles , in which a parallel file is broken into multiple subfiles or segments, typically one per physical processor. The CM 200 supports parallel files, in which each physical processor accesses its own subfile [23] The notion of parallel VP streams is a large scale ....

....file system. We assume the underlying file system manages a parallel file in the traditional manner, as a single stream of bytes. Therefore, Stream can be implemented on top of PFS [11] Thinking Machines CMMD I O [2] and the single stream views of more flexible file systems like PIOUS [19] and Vesta [3] Knowledge of the underlying implementation of the file system (e.g. programmable I O nodes, disk arrays, etc. may be used to optimize the machine specific portions of the Stream run time system, but our discussion is independent of the file system. All of the features of Stream ....

S. A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable High-Performance Computing Conference, pages 71--78, 1994.

....systems. Section 3 describes some simulations and their results and Section 4 describes some measurements on a real system. We summarize our conclusions in Section 5. 2 Background There are many different parallel file systems [Kri94, Pie89, FPD93, Roy93, LIN 93, DdR92, CF94, Dib90, DSE88, MS94, HdC95, HER 95] Most, though not all, are designed for machines that have dedicated I O nodes. Most are based on a fairly traditional Unix like interface, in which individual processes make a request to the file system for each piece of the file they read or write. Increasingly common, ....

Steven A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable High-Performance Computing Conference, pages 71--78, 1994.

....that can be extracted from this work. 2 Related Work In the last years many parallel distributed file systems have been developed and most of them have placed special interest in developing cache strategies and prefetching algorithms (ParfiSys [4] Galley [16] Scotch [11] Zebra [12] PIOUS [24], Vesta [5] and sfs [22] among others) Besides all the above mentioned projects, we would like to place special interest on two other research projects as they have interesting similarities to the work presented here. The first one is the xFS file system [1, 2, 8] xFS is a server less file ....

S. A. Moyer and, V. S. Sunderam "PIOUS: A Scalable Parallel I/O System for Distributed Computing," SHPCC, May 1994, pp.71-79

....streams. Here we extend this work to include automatic mode detection, design details and tradeoffs, and optimizations. Most proposed parallel file systems for MIMD machines give the programmer one view of a file, namely as a single stream of bytes. Some systems, including Vesta [3] PIOUS [21] and others [13] support multifiles , in which a parallel file is broken into multiple subfiles or segments, typically one per physical processor. The CM 200 supports parallel files, in which each physical processor accesses its own subfile [25] The notion of parallel VP streams is a ....

....file system. We assume the underlying file system manages a parallel file in the traditional manner, as a single stream of bytes. Therefore, Stream can be implemented on top of PFS [12] Thinking Machines CMMD I O [2] and the single stream views of more flexible file systems like PIOUS [21] and Vesta [3] Knowledge of the underlying implementation of the file system (e.g. programmable I O nodes, disk arrays, etc. may be used to optimize the machine specific portions of the Stream run time system, but our discussion is independent of the file system. All of the features of Stream ....

S. A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable High-Performance Computing Conference, pages 71--78, 1994.

....window is large enough. From the figure 10, we see that each disk delivered about 3MB s, which is consistent with the sequential read tests on the SP2 disk (table 1) 4 Related Work A number of projects have contributed to the research in scalable I O. Parallel file systems such as [HO93, Mon93, MS94] can run on workstation clusters. Zebra [HO93] has a client server architecture and is implemented on the Sprite operating system, using a log structured file system to store files on the servers. Zebra also implements parity codes in its striping (RAID level 4) but does not address buffering or ....

....Zebra [HO93] has a client server architecture and is implemented on the Sprite operating system, using a log structured file system to store files on the servers. Zebra also implements parity codes in its striping (RAID level 4) but does not address buffering or prefetching issues. The PIOUS [MS94] system uses PVM to communicate over the network, targeting workstation clusters or parallel clusters. PIOUS stripes files and supports a form of transactions (which guarantee file system integrity in case of a crash) and global shared files pointers. MS94] reports results for ethernet tests. ....

[Article contains additional citation context not shown here]

Steven A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable High-Performance Computing Conference, pages 71--78, 1994.

....Related Research PPFS is but one of several new parallel file systems based on user level libraries. The major distinguishing feature of PPFS is its rich interface that supports advertising a variety of access information as well as control and augmentation of input output system policies. PIOUS [18] is a portable input output system designed for use with PVM. PIOUS enforces sequential consistency on file accesses; in contrast, PPFS allows data consistency to be controlled by the application, enabling higher performance as with the genome matching application. PASSION [4] supports outof core ....

Moyer, S. A., and Sundaram, V. S. PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments. In 1994 Scalable High Performance Computing Conference (May 1994), pp. 71--78.

....3) data dependent filtering (Section 4) and working with irregular subsets (Section 5) We present conclusions and a look toward the future in Section 6. 2 Background There are many different parallel file systems [DSE88, Pie89, Dib90, DdR92, Mas92, FPD93, LIN 93, Roy93, CF94, Kri94, MS94, HdC95, HER 95] Most are based on a fairly traditional Unix like interface, in which individual processes make a request to the file system for each piece of the file they read or write. Increasingly common, however, are specialized interfaces to support multidimensional matrices [GL91, ....

Steven A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable High-Performance Computing Conference, pages 71--78, 1994.

....us to track down the access patterns to a particular object among a set of applications and allow us to do optimizations of parallel array object I O. Related work in shared nothing storage system includes IBM s Vesta[5] a scalable parallel I O system at Argonne National Laboratory[23] and PIOUS[25] systems. Our work also falls into the category of shared nothing environments. IBM Vesta puts all the disks on the IO nodes so that their system is nonsymmetrical compared to our system. Vesta has two layer partitions. Programmers not only have the global view of the file but also specify these ....

S. A. Moyer and V. S. Sunderam. PIOUS: A Scalable Parallel I/O Systems for Distributed Computing Environments, Proceedings of IEEE SHPCC, June 1994.

....with each developer s parallel I O interface in order to deal with technical details such as buffering or physical disk layout which clearly should be hidden from the application programming level. A higher level abstraction is provided by several parallel I O libraries like Vesta [6] PIOUS [9], PASSION [5] and B section list B section list B section list B section list D section list E section list B section list and A section list get put get put get put get put b2 b2 b1 b1 b3 b3 b3 b4 b2 b1 b1 time (SSEQ1) time (SSEQ2) time (SSEQ3) time (SSEQ4) time (SSEQ5) S5) S3) S2) S4) ....

S. A. Moyer, V. S. Sunderam. PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments. In Proc. Scalable High Performance Computing Conference, Knoxville, May 1994, pp. 71--78.

.... good approach to take advantage of the inherent parallelism of the MPP, as shown in some parallel file systems (PFS) and I O libraries described in the bibliography (CFS [Pieper89] Vesta [Corbett96] ParFiSys [Carretero96a] PASSION [Choudhary95] Galley [Nieuwejaar96b] Scotch [Gibson95] PIOUS [Moyer94], HFS [Krieger94] Caching can be used in PFS by allocating a buffer cache at the computing nodes and I O nodes. This approach improves I O performance by avoiding unnecessary disk traffic, network traffic and servers load, and also by allowing prefetching and delayed write techniques [Kotz91a, ....

S. A. Moyer and V. S. Sunderam. PIOUS: A Scalable Parallel I/O System for Distributed Computing Environments. In Proceedings of the Scalable HighPerformance Computing Conferece, pages 71--78, 1994.

....they are likely to indicate general trends in file sharing on systems without concurrency control. In this paper we present volatile transactions, a scalable, light weight concurrency control mechanism for parallel file systems. Volatile transactions have been successfully employed in PIOUS [MS94a, MS94b, MS95], a file system for parallel distributed computing environments such as PVM [Sun90, GS92] Performance results demonstrate that sequential consistency is provided without loss of system scalability. Emory University CSTR 950202 2 Volatile Transactions The read inconsistency depicted in Figure 1 ....

Steven A. Moyer and V. S. Sunderam. PIOUS: a scalable parallel I/O system for distributed computing environments. In Proceedings of the Scalable HighPerformance Computing Conference, pages 71--78, 1994. Emory University CSTR-950202

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