N. Nieuwejaar and D. Kotz, A Multiprocessor Extension to the Conventional File System Interface. Technical Report PCS-TR94-230, Dept. Computer Science, Dartmouth College, Sep 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... 15 system advantages disadvantages nCUBE [16] simple partitioning based on bit permutations all sizes must be powers of 2 array partitioning library [7, 6, 23] supports common array partitioning patterns, high level of abstraction must access full array in one operation nested strided [50] supports the common multidimensional access patterns user needs to compute offsets and strides in all dimensions, must access a full multidimensional structure Vesta [13] supports all rectilinear decompositions, simple parameterized interface, express mapping to hardware only 2 D ....

....using the same abstraction (i.e. partitioned arrays) An analogous low level interface has also been proposed recently. It is based on viewing the array data as it is in the file, namely a sequence in some canonical order. Access to a subarray is then expressed as a set of nested strided accesses [50]. This interface requires its user (a programmer or a compiler) to determine the offsets and strides that should be used by the different processes. 3.5 The Vesta Partitioning Scheme All the schemes described so far are based on partitioning the data based on a logical structure as perceived by ....

N. Nieuwejaar and D. Kotz, A Multiprocessor Extension to the Conventional File System Interface. Technical Report PCS-TR94-230, Dept. Computer Science, Dartmouth College, Sep 1994.

....APIs can also occur in a workstation environment. A number of existing file systems provide low level API support for simultaneous, independent, direct access to the multiple disks provided by modern machine architectures. These APIs include extensions to the conventional Unix interface (e.g. NK95] modifications to the conventional Unix interface (e.g. GS95] and other interfaces which differ from Unix (e.g. CFF 95, CFPB93] An I O efficient algorithm would not be easy to program using these low level file system APIs since the programmer must map the high level parallel disk ....

Nils Nieuwejaar and David Kotz. A multiprocessor extension to the conventional file system interface. Technical Report PCS-TR95-253, Dartmouth College, 1995.

.... on the other hand, use larger files and have more sequential access [MK91, GGL93, PP93] Parallel scientific programs access the file with patterns not seen in uniprocessor or distributed system workloads, in particular, complex strided access to discontiguous pieces of the file [KN94, NK94] Finally, scientific applications use files for more than loading raw data and storing results; files are used as scratch space for very large problems as application controlled virtual memory [CK93] In short, multiprocessors need new file systems that are designed for parallel scientific ....

Nils Nieuwejaar and David Kotz. A multiprocessor extension to the conventional file system interface. Technical Report PCS-TR94-230, Dept. of Computer Science, Dartmouth College, September 1994.

.... Scientific applications use larger files and have more sequential access [MK91, GGL93, PP93] Parallel scientific programs access the file with patterns not seen in uniprocessor or distributed system workloads, in particular, complex strided access to discontiguous pieces of the file [KN94, NK94] Finally, scientific applications use files for more than loading raw data and storing results; files are used as scratch space for very large problems as application controlled virtual memory [CK93] In short, multiprocessors need new file systems that are designed for parallel scientific ....

....Collective writes are similar. We assume that the request can be specified compactly (perhaps by a compiler or library) with a few parameters describing a regular mapping of file data to CP buffers (for example, the out of core array descriptors of [TBC94] or the strided requests of [NK94] Even sending several KBytes of description information to the IOPs would add negligible overhead to the cost of a large I O. A note about the barriers in Figure 2c. The cost of the barriers themselves is negligible compared to the time needed for a large I O transfer. Of more concern is the ....

[Article contains additional citation context not shown here]

Nils Nieuwejaar and David Kotz. A multiprocessor extension to the conventional file system interface. Technical Report PCS-TR94-230, Dept. of Computer Science, Dartmouth College, September 1994.

.... Scientific applications use larger files and have more sequential access [MK91, GGL93, PP93] Parallel scientific programs access the file with patterns not seen in uniprocessor or distributed system workloads, in particular, complex strided access to discontiguous pieces of the file [KN94, NK94] Finally, scientific applications use files for more than loading raw data and storing results; files are used as scratch space for very large problems as application controlled virtual memory [CK93] In short, multiprocessors need new file systems that are designed for parallel scientific ....

Nils Nieuwejaar and David Kotz. A multiprocessor extension to the conventional file system interface. Technical Report PCS-TR94-230, Dept. of Computer Science, Dartmouth College, September 1994.

.... both cases, applications accessed large files (megabytes or gigabytes in size) using surprisingly small requests (on the Intel, 96 of read requests were for less than 200 bytes) On further examination, we discovered that most of the files were accessed in complex yet highly regular patterns [NK94] most likely due to accessing multidimensional matrices. Interfaces. Most parallel file systems present the traditional abstraction of a file as a sequence of bytes with Unix interface semantics, and add a few extensions to control the behavior of an implicit file pointer shared among the ....

....file pointer shared among the processes. This low level interface, which restricts each request to a contiguous portion of the file, is one reason for the predominance of small requests found by the CHARISMA project. Higher level interfaces, such as specifying a strided series of requests [NK94, Cra94] or accessing data through a mapping function [CF94, DdR92, Kot93] provide valuable semantic information to the file system, which can then be used for optimization purposes. Interfaces that allow the programmer to express collective I O activity, in which all processes cooperate to make a ....

Nils Nieuwejaar and David Kotz. A multiprocessor extension to the conventional file system interface. Technical Report PCS-TR94-230, Dept. of Computer Science, Dartmouth College, September 1994.

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