Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....this problem by introducing file modes. The different modes specify the semantics of simultaneous operations by multiple processes. Once a mode is defined, conventional read and write operations are used to access the data, and their semantics are determined by the mode. The most common modes are [3, 17, 18, 27]: mode description examples broadcast all processes collectively Express singl reduce access the same data PFS global mode CMMD sync broadcast scatter all processes collectively Express multi gather access a sequence of data CFS modes 2 and 3 blocks, in rank order PFS sync record CMMD ....

.... use other than mode 0 [20] and that various simple partitioning schemes do indeed account for 99 of observed parallel I O patterns [25] In addition to the commercial offerings (IBM SP2 PIOFS [7] Intel iPSC CFS [26, 28] Intel Paragon PFS [11, 30] Ncube [9] and Thinking Machines CM 5 sfs [3, 21]) there has been a recent flurry of activity in the research community. PIOUS [23, 24] and PETSc Chameleon I O [13] are both widely available non proprietary portable parallel I O interfaces. PIOUS is a PVM based parallel file interface. Files can be declustered across disks in a round robin ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....control the distribution of data blocks on physical disks, and easily present a variety of access patterns with a minimum number of I O requests. However, the majority of current file systems for parallel computers are based on a linear file model and do not support user controllable features [1, 7, 9]. The Vesta parallel file system [3] supports a twodimensional file abstraction that provides a flexible way of data partitioning for parallel access, but it does not support the partitioning multidimensional data structures. The Galley [8] file system allows users to control declustering data ....

M. L. Best, A. Greenberg, C. Stanfill, and L. W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....modes specify the semantics of simultaneous operations by multiple processes. Once a mode is defined, conventional read and write operations MPI IO Parallel I O Interface 3 are used to access the data, and their semantics are determined by the mode. The most common modes are the following [2, 13, 16, 17, 26]: mode description examples broadcast all processes collectively Express singl reduce access the same data PFS global mode CMMD sync broadcast scatter all processes collectively Express multi gather access a sequence of data CFS modes 2 and 3 blocks, in rank order PFS sync record CMMD ....

.... software [8] Recent studies show that various simple partitioning schemes do indeed 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 ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....Further, if each VP explicitly accesses its own subfile, many fine grained file operations are required. Performance suffers significantly when this approach is used. Modes were introduced with the earliest parallel file systems [20, 25] and some more recent systems have adopted this approach [2, 11]. The systems provide a limited number of operations whose synchronization requirements and file access semantics vary depending on the current mode. The implementations of modes on these systems have several drawbacks. They are machine dependent. They give a single file operation multiple ....

....Stream implementation, while a few routines make calls to the machine dependent 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 ....

M. L. Best, A. Greenberg, C. Stanfill, and L. W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....virtual machine. Further, if each VP explicitly accesses its own subfile, many finegrained file operations are required. Performance suffers significantly when this approach is used. Modes were introduced with the earliest parallel file systems [22, 27] and some more recent systems use modes [2, 12]. The systems provide a limited number of operations whose synchronization requirements and file access semantics vary depending on the current mode. The implementations of modes on these systems have several drawbacks. They are machine dependent. They give a single file operation multiple ....

....while the functions are executed. The computation performed depends on the context when the function is called. The following examples include a prototype for function foo and a call to foo from within a where statement. double:S foo(double:S x, double:S y) with (S) where (x 0) y[2] = foo(y[0] y[1] The second type of function with which parallel data can be manipulated is an elemental function. Elemental functions are an important extension to C formulated by ANSI X3J11.1 [8] Elemental functions allow VPs to invoke scalar functions conceptually by passing individual ....

[Article contains additional citation context not shown here]

M. L. Best, A. Greenberg, C. Stanfill, and L. W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....the simple Unix like interface has worked well in the past, it is clear that it is not well suited to parallel applications, which have more complicated access patterns. One extension to the conventional interface offered by several multiprocessor file systems is a shared file pointer [Pie89, BGST93] This provides a mechanism for regulating access to a shared file by multiple processes in a single application. The simplest shared file pointer is one which supports an atomic append mode (as in [LMKQ89] page 174) Intel s CFS provides this in addition to several more structured access modes ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....For example, many exhibit forward jumping sequential patterns [24, 31] many of which are actually complex strided patterns [26, 27] Clearly, parallel file systems must be redesigned to fit these common access patterns. Several recent works have proposed changes to the file system interface [3, 7, 8, 10, 12, 18, 19, 26]. One such proposed interface is collective I O. In a traditional file system interface, processes within a parallel job often have to express the transfer of a large object (e.g. a large matrix) as small, non contiguous, per processor requests, thereby losing valuable semantic information that a ....

.... Advances in Parallel I O Earlier generations of parallel file systems like Intel CFS extended the Unix interface with file pointer modes for parallel access [5, 28] The Scalable File System (SFS) on the CM 5 also provides a Unixlike interface, but has an additional collective I O interface [3]. Among more recent parallel file systems, Vesta [8] allows users to specify both the logical partitioning of file data among the CPs and the physical partitioning of file blocks across the disks. Some recent proposals focus on devising and implementing new and meaningful interfaces. Corbett et ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....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 a single coordinated ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....the file, like the second. Each call specifies the bytes of the file that should be transferred. This interface is common when using the C programming language in most MIMD systems, although many have special file pointer modes that help in a few simple situations (Intel CFS [Pie89] and TMC CMMD [BGST93] for example) None of these allow the processor to make a single file system request for a complex distribution pattern. More sophisticated interfaces, such as the nested batched interface [NK96b] can specify a list, or a strided series, of transfers in a single request. This latter interface ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....related work, and finally, in Section 6, we conclude and describe our future plans. 2 Background 2. 1 Parallel File Systems Most existing multiprocessor file systems are based on the conventional Unix like file system interface in which files are seen as an addressable, linear stream of bytes [BGST93, Pie89, LIN 93, WMR 94] To provide higher throughput, the file system typically declusters files (i.e. scatters the blocks of each file across multiple disks) thus allowing parallel access to the file, reducing the effect of the bottleneck imposed by the relatively slow disk speed. ....

....an implicit file pointer. One enhancement to the conventional interface, which is offered by several multiprocessor file systems, is a file pointer that is shared among the processes in an application and provides a mechanism for regulating access to a shared file by those processes [Pie89, BGST93] The simplest shared file pointer is one which supports an atomic append mode (as in [LMKQ89] page 174) Most parallel file systems provide this mode in addition to several more structured access modes (e.g. round robin access to the file pointer) We compare Galley to other, more ....

[Article contains additional citation context not shown here]

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....was supported in part by the NASA Ames Research Center under Agreement Number NCC 2 849. 2 The Conventional Interface Many existing multiprocessor file systems are based on the conventional Unix like file system interface in which files are seen as an addressable, linear stream of bytes [BGST93, Pie89, LIN 93] To provide higher throughput, the file system typically declusters files (i.e. scatters the blocks of each file across multiple disks) thus allowing parallel access to the file, reducing the effect of the bottleneck imposed by the relatively slow disk speed. Although the ....

....in a parallel environment. While our focus in this paper is the improvement of the interface to a linear file model, the enhancement or outright replacement of that model is worthy of further investigation. One common enhancement to the conventional interface is a shared file pointer [Pie89, RP95, BGST93, FBD94] which provides a mechanism for regulating access to a shared file by multiple processes in a single application. The simplest shared file pointer is one which supports an atomic append mode (as in [LMKQ89] page 174) Intel s CFS provides this mode in addition to several more structured ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....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 these requests as a single ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 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 Intel CFS and PFS [19] IBM Vesta [5] TMC [1], and HFS [15] to name a few. There are also several systems intended for workstation clusters, such as PIOUS [16] and VIP FS [10] All of these systems decluster file data across many disks to provide parallel access to the data of any file. A full characterization of these systems is not ....

M. L. Best, A. Greenberg, C. Stanfill, and L. W. Tucker. CMMD I/O: A parallel Unix I/O. In Proc. of the 7th IPPS, pp. 489--495, 1993.

....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 a single coordinated ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....second. Each call specifies the bytes of the file that should be transferred. This interface is common when using the C programming language in most MIMD systems, although many have special file pointer modes that help in a few simple situations (Intel CFS [Pie89] Intel PFS [RP95] and TMC CMMD [BGST93] for example) None of these allow the processor to make a single file system request for a complex distribution pattern. More sophisticated interfaces, such as the nested batched interface [NK95] can specify a list, or a strided series, of transfers in a single request. This latter interface ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....patterns [NK95, NKP 95] Files are much larger and file lifetimes much longer than in conventional Unix workloads [PEK 95] Clearly, parallel file systems must be redesigned to fit these common access patterns. Several recent works have proposed changes to the file system interface [BGST93, CFF 95, CFPB93, dBC93, GGL93, Kot93, Kot94, NK95] One such proposed interface is collective I O. In a traditional file system interface, processes within a parallel job often have to express the transfer of a large object (e.g. a large matrix) as small, noncontiguous, per processor ....

.... Advances in Parallel I O Earlier generations of parallel file systems like Intel CFS extended the Unix interface with file pointer modes for parallel access [Nit92] The Scalable File System (SFS) on the CM 5 also provides a Unix like interface, but has an additional collective I O interface [BGST93] Among more recent parallel file systems, Vesta [CFPB93] breaks away from the tradition of a one dimensional file structure, and allows files to be viewed as two dimensional entities where the user is capable of specifying both the logical partitioning of file data among the CPs and the physical ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....of libraries. ffl Scale to many compute and I O processors. ffl Minimize memory and performance overhead. 3 File Structure Most existing multiprocessor file systems are based on the conventional Unix like file system interface in which a file is seen as an addressable, linear sequence of bytes [BGST93, Pie89, LIN 93] The file system typically declusters files (i.e. scatters the blocks of each file across multiple disks) allowing parallel access to the file. This parallel access reduces the effect of the bottleneck imposed by the relatively slow disk speed. Although the file is actually ....

....actually deleted until it is closed by every CP that has it opened. For convenience, there are gfs all open, gfs all close, and gfs all delete calls as well. 5. 3 Data Operations Most parallel file systems present applications with an application interface similar to that of Unix [Pie89, RP95, BGST93] While this interface is simple and familiar to programmers, it was not designed to allow parallel applications to access parallel disks. In particular, it does not allow programmers to issue the highly structured requests that we have observed to be common among parallel, scientific ....

[Article contains additional citation context not shown here]

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 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 ....

M. L. Best, A. Greenberg, C. Stanfill, and L. W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....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 processor ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....in libraries, ffl allow easy and efficient implementations of libraries, ffl be scalable enough to run well on multiprocessors with tens or hundreds of nodes, ffl minimize memory and performance overhead. 5. 2 File Structure Most parallel file systems are based on a Unix like, linear file model [BGST93, LIN 93, Pie89] Under this model, a file is seen as an addressable, linear sequence of bytes. Applications can issue requests to read or write contiguous subranges of that sequence of bytes. A parallel file system typically declusters the data within those files (i.e. scatters the blocks ....

....sfs was designed as a low level file system. Users were expected to access it through the I O support built in to one of their data parallel languages or through the CMMD library, described below. 8.1. 3 CMMD The CMMD library allows CM 5 applications to be written in a control parallel style [BGST93] CMMD maintains the traditional stream of bytes abstraction of a file and supports all the standard Unix like operations, which may be executed by each compute node individually. In addition to the standard operations, CMMD includes support for parallel applications in the form of access modes ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....run well on multiprocessors with dozens or hundreds of nodes. ffl Minimize memory and performance overhead. Galley is targeted at distributed memory, MIMD machines such as IBM s SP 2 or Intel s Paragon. 3 File Structure Most existing multiprocessor file systems are based on a Unix like model [BGST93, Pie89, LIN 93] Under this model, a file is seen as an addressable, linear sequence of bytes. Applications can issue requests to read or write data contiguous subranges of that sequence of bytes. A parallel file system typically declusters files (i.e. scatters the blocks of each file across ....

....CFS and PFS provide several modes, each of which provides the applications with a different set of semantics governing how the file pointers are shared. Other multiprocessor file systems with this style of interface are SUNMOS and its successor, PUMA [WMR 94] sfs [LIN 93] and CMMD [BGST93] Like the systems mentioned above, PPFS provides the end user with a linear file that is accessed with primitives that are 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 maps ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....process. In the New Mexico Order, the data array is striped across the four computation nodes in units 51 . A[113 114] A[1 2] A[15 16] Disk 0 Disk 1 Disk 2 Disk 3 Disk 4 Disk 5 Disk 6 Disk 7 A[7 8] A[21 22] A[5 6] A[19 20] A[3 4] A[17 18] A[115 116] A[119 120] A[127 128] A[117 118] A[9 10] A[23 24] A[121 122] A[11 12] A[25 26] A[123 124] A[13 14] A[27 28] A[125 126] Parity Parity Parity Parity Figure 8.3: Data Layout on the Disk Array of 14 data elements (i.e. the blocking factor) Note that ....

....portions from each computation node are striped across the disks in units of 16 bytes, the size of the payload of a data network packet. In other words, the first data packet transfers data elements A[1] and A[2] from node zero to disk zero, the second data packet transfers data elements A[3] and A[4] from node zero to disk one, and so on. File reads are the exact opposite of file writes; the array is first read from the disks into the New Mexico Order canonical layout illustrated in Figure 8.2. Then, the data is reordered back into the user distribution shown in Figure 8.1. 52 File Format ....

[Article contains additional citation context not shown here]

Best, M. L., Greenberg, A., Stanfill, C., and Tucker, L. W. CMMD I/O: A Parallel Unix I/O. In Proceedings of the 1993 International Parallel Processing Symposium (1993), pp. 489--495.

....2.2 Existing Parallel File Systems Existing parallel I O models are often closely tied to the machine architecture as well as to the programming model. Typically jobs can access files in different I O modes , which determine how a file pointer is shared among clients running in individual nodes [7, 4, 18]. The HFS [20] and KSR1 [17] file systems use a memory mapped interface. The nCUBE [9] and Vesta [5] allow more user control over data layout by providing per process logical views of the data. In PIFS [11] the file system controls which processor handles which part of the file to exploit memory ....

....arrays in SDA files via either special library calls or normal Fortran READ and WRITE statements. CMMD allows multiple threads of control, one for each PN. CMMD I O provides a variety of I O modes in some, action is taken by a single PN; in others, all PN s co operatively perform parallel I O [4]. 3 Tracing Methodology The 512 node NCSA CM 5 is generally it divided into 5 partitions of size 32, 32, 64, 128 and 256 nodes; at times the machine is reconfigured as a single 512 node partition. The SDA has 118 data disks and 1 parity disk for a total capacity of about 138 Gbytes. A single ....

M. L. Best, A. Greenberg, C. Stanfill, and L. W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....2.2 Existing Parallel File Systems Existing parallel I O models are often closely tied to the machine architecture as well as to the programming model. Typically jobs can access files in different I O modes , which determine how a file pointer is shared among clients running on individual nodes [7, 4, 18, 14, 22]. The Hurricane [20] and KSR1 [17] file systems use a memorymapped interface. The nCUBE [9] and Vesta [5] file systems allow more user control over data layout by providing per process logical views of the data. In PIFS (Bridge) 11] the file system controls which processor handles which part of ....

....arrays in SDA files via either special library calls or normal Fortran READ and WRITE statements. CMMD allows multiple threads of control, one for each PN. CMMD I O provides a variety of I O modes in some, action is taken by a single PN; in others, all PNs co operatively perform parallel I O [4]. 3 Tracing Methodology The 512 node NCSA CM 5 is generally divided into 5 partitions of size 32, 32, 64, 128 and 256 nodes; at times the machine is reconfigured as a single 512 node partition. The SDA has 118 data disks and 1 parity disk for a total capacity of about 138 Gbytes. A single file ....

M. L. Best, A. Greenberg, C. Stanfill, and L. W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

....programming model. Nonetheless, 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) ....

....programming languages like C, C , and f77. Under CMMD the user sees multiple threads of control, one for each PN. CMMD I O (again layered on top of SFS) provides a variety of I O modes in some, action is taken by a single PN; in others, all PN s co operatively perform a parallel I O [Thi93c, BGST93] 3 Tracing Methodology The CM 5 at the National Center for Supercomputing Applications (NCSA)was chosen as our target machine because this is one of the most widely used CM 5 machines in the United States. The user population is distributed all across the nation and there are approximately 1000 ....

Michael L. Best, Adam Greenberg, Craig Stanfill, and Lewis W. Tucker. CMMD I/O: A parallel Unix I/O. In Proceedings of the Seventh International Parallel Processing Symposium, pages 489--495, 1993.

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