P. J. Roy, "Unix file access and caching in a multicomputer environment". In USENIX Mach III Symp., pp. 21--37, Apr 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... with actual application I O for 29 limited buffer space, especially in installations that had a high ratio of compute nodes to I O nodes [51] This was the reason for streaming data through the I O nodes without any buffering in the Intel PFS design (which is based on the OSF 1 AD file system [58]) Explicit support for collective I O operations can also be combined with prefetching and buffer management. This is done in TIP (Transparent Informed prefetching) 55] The idea is to provide the system with hints that inform it about future reference patterns. The system can then decide what ....

P. J. Roy, "Unix file access and caching in a multicomputer environment". In USENIX Mach III Symp., pp. 21--37, Apr 1993.

....basis. Mode 2: Reads and writes are treated as global operations and a global synchronization is performed. Mode 3: A synchronous ordered mode is provided, but all write operations have to be of the same size. IV. Parallel Operating Systems 195 The PFS (Parallel File System) [Roy93] is a striped file system which stripes files (but not directories or links) across the UNIX file systems that were assembled to create the file system. The size of each stripe, the stripe unit is determined by the system administrator. Contrary to the previous two file systems, the file system ....

Roy, P.J., Unix file access and caching in a multicomputer environment, Proc. the Usenix Mach III Symposium, 1993, 21-37.

....background information about multiprocessor file 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 ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....across many disks, which are accessed in parallel. Most extend a traditional file abstraction (a growable, addressable, linear sequence of bytes) with some parallel file access methods. The most common provide I O modes that specify whether and how parallel processes share a common file pointer [14, 21, 22, 23, 24, 25]. Some systems are based on a memory mapped interface [26, 27] and two provide a way for the user to specify per process logical views of the file [28, 29] Some provide SIMDstyle transfers [30, 31, 25, 18] Finally, in addition to shared file pointers, MPI IO allows applications to describe a ....

Paul J. Roy, "Unix file access and caching in a multicomputer environment", in Proceedings of the Usenix Mach III Symposium, 1993, pp. 21--37.

.... of workstation, a distributed operating system is needed that offers all services required by existing software, new services for distributed systems (e.g. file sharing) and additional, efficient services for parallel systems (e.g. load balancing) Current distributed operating systems (e.g. [1, 2, 3, 4]) try to approach theses goals by extending existing operating systems. State of the art operating systems are build on the clientserver model [5] Based on a micro kernel the main functionality of the operating system is provided by a group of cooperating processes, called servers, that is used ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....distribution (Section 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 ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....file system call is needed for each contiguous chunk of the file. The file system is thus faced with concurrent small requests from many processors, instead of the single large request that would have occurred on a uniprocessor. Indeed, since most multiprocessor file systems [CF94, FPD93, Pie89, Roy93, DdR92, LIN 93, BGST93, Dib90, DSE88] decluster file data across many disks, each application request may be broken into even smaller requests that are sent to different IOPs. It is difficult for the file system, which is distributed across many I O processors, to recognize these requests as ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....Disks (PID) in an MIMD multiprocessor, with separate compute processors (CP) and I O processors (IOP) concurrent small requests from many processors, instead of the single large request that would have occurred on a uniprocessor. Indeed, since most multiprocessor file systems [CF94, FPD93, Pie89, Roy93, DdR92, LIN 93, BGST93, Kri94, Dib90, DSE88, FBD94] decluster file data across many disks, each application request may be broken into even smaller requests that are sent to different IOPs. It is difficult for the file system, which is distributed across many I O processors, to recognize ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....flexibility and performance is a continual challenge. Recent systems software, both academic and commercial, is reaching for more flexibility by moving system components into separate modules, which are often loaded in separate address spaces. For example, Windows NT [Custer 93] and OSF 1 MK [Roy 93] have moved their application programmer interfaces (Win32 and Unix, respectively) into address spaces outside the kernel. The X Window System [Scheifler Gettys 90] places the display manager and the window manager functionality in their own address spaces. This trend increases the importance ....

Paul J. Roy. Unix File Access and Caching in a Multicomputer Environment. In Proceedings of the USENIX Mach III Symposium, pages 21-38, April 1993.

....Disks (PID) in an MIMD multiprocessor, with separate compute processors (CP) and I O processors (IOP) concurrent small requests from many processors, instead of the single large request that would have occurred on a uniprocessor. Indeed, since most multiprocessor file systems [CF94, FPD93, Pie89, Roy93, DdR92, LIN 93, BGST93, Dib90, DSE88] decluster file data across many disks, each application request may be broken into even smaller requests that are sent to different IOPs. It is difficult for the file system, which is distributed across many I O processors, to recognize these requests as ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....network. Nodes with attached disks are usually reserved as I O nodes, while applications run on some cluster of the remaining compute nodes. In the past few years, many parallel file systems have been described in the literature, including Bridge PFS [Dib90] CFS [Pie89] nCUBE [DdR92] OSF PFS [Roy93] sfs [LIN 93] Vesta PIOFS [CFP 95] HFS [KS96] PIOUS [MS94] RAMA [MK95] PPFS [HER 95] Scotch [GSC 95] and Galley [NK96a, NK96b] Many more techniques for improving the performance of parallel file systems have been described, including caching and prefetching [KE93b, KE93a, ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....the fundamental idea of declustering file blocks across many disks for parallel access has been a common feature. Typically jobs can access files in different I O modes , which determine how a file pointer is shared among clients running in individual nodes [Cro89, BGST93, Kot93, Pie89, Roy93] HFS [KS93] and the KSR1 [Ken92] file system, use a memory mapped interface. On the nCUBE [DdR92] and in Vesta [CFPB93] the user has more control over data layout. These systems provide a per process logical view of the data. In PIFS (Parallel Interleaved File System) Dib90] the file system ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....of a file across many disks, which are accessed in parallel. Most extend a traditional file abstraction (a growable, addressable sequence of bytes) with some parallel file access methods. The most common provide I O modes that specify whether and how parallel processes share a file pointer [7, 30, 33, 2, 17]. Some are based on a memorymapped interface [23, 22] Some provide a way for the user to specify per process logical views of the file [5, 9] Some provide SIMD style transfers [34, 24, 16] PIFS (Bridge) 11] allows the file system to control which processor handles which parts of the file, to ....

P. J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....the CFS file system for the iPSC 2 and iPSC 860 [105] implement the Unix I O interface in a user level library on top of a lower level internal system call interface. The OSF 1 file system for the Paragon implements the Unix I O interface in an emulation library on top of a mapped file interface [111]. Both mapped file I O and Unix I O interfaces are exported to the application on the Paragon. Parallel application level I O interfaces Recently, there have been several proposals for developing application level parallel I O libraries [9, 25, 36, 44, 49, 119] By providing an interface that ....

....sequentially according to the position of the data on the disk. It is also important to consider where in memory data should be cached. The Intel CFS prefetches data sequentially into memory on the I O nodes. On the other hand, the Intel Paragon prefetches data into the memory of the compute nodes [111]. With a shared memory multiprocessor there is a greater range of alternatives. For example, data can be read into a remote memory [78] and, rather than ejecting data from the file cache, it can be migrated to remote memories. 2.5.3 Disk block distribution The majority of file systems for ....

P. J. Roy. UNIX file access and caching in a multicomputer environment. In USENIX Mach III symposium proceedings, pages 21--37, April 1993.

....across many disks, which are accessed in parallel. Most extend a traditional file abstraction (a growable, addressable sequence of bytes) with some parallel file access methods. The most common provide I O modes that specify whether and how parallel processes share a file pointer [Cro89, Pie89, Roy93, BGST93, Kot93] Some are based on a memory mapped interface [KSR92, KS93] Some provide a way for the user to specify per process logical views of the file [CFPB93, DdR92] Some provide SIMD style transfers [TMC87, Mas92, GGL93] PIFS (Bridge) Dib90] allows the file system to control which ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

.... the full power of the Unix operating system to each compute node [ZRB 93] This power does come at some cost; on a Paragon, SUNMOS requires only 256KB of memory on each node, but OSF 1 AD occupies nearly 10MB [MMRW94] File system access in OSF 1 is built on top of Mach Memory Objects [Roy93] Since memory objects are restricted to multiples of the page size, small, non contiguous requests, such as those common in parallel scientific workloads, may lead to poor performance. There is no support for a multiprocessing environment at the user interface. Furthermore, the designers ....

....those common in parallel scientific workloads, may lead to poor performance. There is no support for a multiprocessing environment at the user interface. Furthermore, the designers explicitly did not intend for OSF 1 to satisfy the I O performance needs of scientific, supercomputer applications [Roy93] 8.1.7 PPFS Like the systems mentioned above, PPFS provides the end user with a linear file that is accessed with primitives similar to the traditional read( write( interface [HER 95] In PPFS, however, the basic transfer unit is an application defined record, rather than a byte. PPFS ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....high speed network. Nodes with attached disks are usually reserved as I O nodes, while applications run on some cluster of the remaining compute nodes. In the past few years, many parallel file systems have been described in the literature, including Bridge PFS [12] CFS [35] nCUBE [9] OSF PFS [38], sfs [27] Vesta PIOFS [6] HFS [25] PIOUS [30] RAMA [29] PPFS [19] Scotch [15] and Galley [31, 32] Many more techniques for improving the performance of parallel file systems have been described, including caching and prefetching [24, 23, 34] two phase I O [10] disk directed I O [20] ....

P. J. Roy. Unix file access and caching in a multicomputer environment. In Proc. of the Usenix Mach III Symposium, pages 21--37, 1993.

....workstations, like Solaris and HP UX, the first steps are taken. All of them support network file systems (NFS) or network information services (NIS) Additional services are added by servers like distributed shared memory servers (DSM) or distributed light weight processes (activities) servers [1, 2, 3, 4]. The remaining question is which services should be added to these servers to reach the goal of a distributed resource management. Many strategies are known for mapping or for load balancing. Also many different strategies for distributed shared memory are developed. But not all strategies are ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....ways of sharing a common file pointer. Unfortunately, caching and prefetching are completely out of the control of the user, and the pattern for declustering the file across disks is not predictable and mostly out of the user s control. The Parallel File System (PFS) for the Intel Paragon [Roy93] supports our list of capabilities [Rul93] with a few restrictions. While it appears that caching can be disabled on the compute node, I O node caching is always active. Declustering parameters are determined on a per filesystem basis. Finally, the Paragon does not maintain parity across I O ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

....the results. We conclude with commentary on the advantages and disadvantages of high level, collective requests, and on the underlying technique of disk directed I O. 2 Background File systems. There are many parallel file systems today, including Bridge [DSE88] Intel CFS [Pie89] Intel PFS [Roy93] IBM Vesta [CF94] nCUBE [DdR92] TMC sfs [LIN 93, BGST93] Hurricane File System [Kri94] and SPIFFI [FBD95] There are also several systems intended for workload clusters, such as PIOUS [MS94] and VIP FS [dHC94] All of these systems decluster file data across many disks to provide ....

Paul J. Roy. Unix file access and caching in a multicomputer environment. In Proceedings of the Usenix Mach III Symposium, pages 21--37, 1993.

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