Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 Scalable File System. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, Manchester, England, 1994. ACM Press.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....or multiprocessor. Local le systems designed for message passing multiprocessors do not share memory transparently between processors but use the message passing model to explicitly communicate and share data at a coarse granularity. Examples of such le systems include the Thinking Machines sfs [38], Galley File System [48] IBM Vesta PIOFS File System [17] IBM GPFS [67] Intel PFS [79] Intel CFS [10] and PVFS [13] Local le systems designed for shared memory multiprocessors share data structures at a very ne level of granularity. Examples of such le systems include SGI XFS [75] IBM ....

....results, to support the claims. There have been many performance studies conducted on le systems for message passing multiprocessors, multicomputers, and clusters such as the performance of the Andrew File System [27] Sprite Network File System [47] Thinking Machines Scalable File System [38], Galley Parallel File System [48, 49, 50] IBM Vesta Parallel File System [17] IBM General Parallel File System [25] IBM Parallel I O File System [79] Intel Parallel File System [79] Intel Concurrent File System [10] and Parallel Virtual File System [13] However, there have been few ....

Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 Scalable File System. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156-165, Manchester, England, 1994. ACM Press.

....of the CM 5 machine. has two disks attached to it. The CM 5 machine supports several different file systems. We are interested in the Scalable File System (SFS) 58] for the following reasons. First, the SFS supports several parallel I O interfaces and can provide high bandwidth up to 95 MB s [53]. Second, the SFS is based on the scalable disk array (SDA) which is a collection of high speed commodity disks connected to the CM 5 I O node. A SFS file is always available in a serial order, however, it is physically striped across disks in unit of sixteen bytes. Our program accesses the SFS ....

T. T. Kwan and D. A. Reed. Performance of the CM-5 scalable file system. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, July 1994.

....data to be accessed from a give file offset. Many file systems provide extensions for expressing user level I O parallelism. File access modes provided in Intel machines (iPSC 860, Touchstone Delta, and Paragon) and in the Thinking Machine CM 5 are examples of such extensions [BCR93, KN93, ACR95, KR94] Recently, several research groups have proposed low level file system interfaces that support irregular file access patterns. The nested batched interface supported by the Galley file system [Nie96] provides a compact representation of irregular file access patterns. The Scalable I O low level ....

Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 Scalable File System. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, July 1994.

....on today s multiprocessors. When they find a multiprocessor that is configured with sufficient parallel I O hardware (unfortunately, many are not) they often discover that the file system software is not designed to meet their needs [CK93, KN94, PEK 95] or has poor performance [Nit92, KR94, FBD94] As a result, there are several proposals for new interfaces, run time libraries, compilers, languages, and file systems to support parallel applications on parallel computers. The focus of this paper is on a file system technique called disk directed I O, which can dramatically improve ....

Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 scalable file system. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, July 1994.

....of data into larger messages for transferring between CPs and IOPs. 4 Performance Performance studies of parallel file systems tend to focus on the performance of large, sequential requests. Indeed, most do not even examine the performance of requests of fewer than many kilobytes [Nit92, BBH95, KR94] As discussed above, recent workload characterizations show that parallel file systems are frequently called upon to service many small requests. This disparity means that most performance studies actually fail to examine how a file system can be expected to perform when running real ....

Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 scalable file system. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, July 1994.

....have also studied the performance of parallel file systems. Bordawekar et al. 4] performed a detailed performance evaluation of the Concurrent 2 See http: www.cacr.caltech.edu SIO for information on the Scalable I O Initiative File System (CFS) on the Intel Touchstone Delta. Kwan and Reed [15] measured the performance of the CM 5 Scalable File System. Feitelson et al. 10] studied the performance of the Vesta file system. Nieuwejaar and Kotz [16] present performance results for the Galley parallel file system. Several researchers have measured the performance of the Concurrent File ....

T. Kwan and D. Reed. Performance of the CM-5 Scalable File System. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, July 1994.

....4e 06 6e 06 8e 06 1e 07 1.2e 07 Rate in bytes sec 32 processor reads 32 processor writes Figure 10: INPUT AND OUTPUT PERFORMANCE of application output and actual input rates on the Maryland CM 5. Notice the performance drop for both input and output rates beyond 6MB 5 Related work Kwan and Reed [6] study the performance of the Scalable File System on the NCSA CM 5. Their motivation was to compare the performance of the CM 5 parallel I O interfaces of CM Fortran and C CMMD. Their C CMMD experiments are similar to ours. Our results on the CM 5 are preliminary but are comparable with theirs. ....

....and Choudhary [7] present an experimental performance evaluation of the Touchstone Delta Concurrent File System. The actual read (6 MB sec) and actual write (1. 2 MB sec) rates we measured on the IU Paragon are a lot lower than the rates they report for the Intel Touchstone Delta (also reported in [6]) 6 Conclusions We identify two important data transfer rates for I O: the application rate and the actual rate. The ap 0 1e 06 2e 06 3e 06 4e 06 Amount of data written by each processor (n p) in bytes 0 2e 06 4e 06 6e 06 8e 06 32 processor paragon 32 processor CM 5 Figure 11: COMPARING THE ....

Thomas Kwan and Daniel Reed. Performance of the CM-5 scalable file system. In Proceedings of International Conference on Supercomputing 94, July 1994. Also accessible on the World Wide Web at http://bugle.cs.uiuc.edu/.

....section of the code be modified. Chapter 6 Performance of Galley Most studies of multiprocessor file systems have focused primarily on the systems performance on large, sequential requests. Indeed, most do not even examine the performance of requests of fewer than many kilobytes [Nit92, BBH95, KR94] As discussed earlier, multiprocessor filesystem workloads frequently include many small requests. This disparity between the observed and benchmarked workloads means that most performance studies actually fail to examine how a file system can be expected to perform when running real ....

Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 scalable file system. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, July 1994.

....messages for transferring between CPs and IOPs. 6 Performance Most studies of multiprocessor file systems have focussed primarily on the systems performance on large, sequential requests. Indeed, most do not even examine the performance of requests of fewer than many kilobytes [Nit92, BBH95, KR94] As discussed above, multiprocessor file system workloads frequently include many small requests. The disparity between the measured and benchmarked workloads means that most performance studies actually fail to examine how a file system can be expected to perform when running real applications ....

Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 scalable file system. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, July 1994.

....of data into larger messages for transfer between CPs and IOPs. 6 Performance Most studies of multiprocessor file systems have focused primarily on the systems performance on large, sequential requests. Indeed, most do not even examine the performance of requests of fewer than many kilobytes [22,2,14]. As discussed earlier, multiprocessor file system workloads frequently include many small requests. This disparity between the observed and benchmarked workloads means that most performance studies actually fail to examine how a file system can be expected to perform when running real ....

Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 scalable file system. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, July 1994.

....file system, developed by IBM, allows applications to define logical partitions, and some access information. Applications can also exercise some control over the data distribution. Related work also includes a number of commercial parallel file systems the CM 5 Scalable Parallel File System [19, 18], the Intel Concurrent File System [10] for the iPSC 2 and iPSC 860, and the Intel Paragon s Parallel File System [14] These provide data striping and a small set of parallel file access modes. In many cases, these access modes do not allow the application enough control to extract good ....

Kwan, T. T., and Reed, D. A. Performance of the CM-5 Scalable File System. In Proceedings of the 1994 ACM International Conference on Supercomputing (July 1994).

....parallel file system [5] allows applications to define logical partitions, data distributions, and some access information. However applications have little control over caching. Related work also includes a number of commercial parallel file systems the CM 5 Scalable Parallel File System [17, 16], the Intel Concurrent File System [9] for the iPSC 2 and iPSC 860, and the Intel Paragon s Parallel File System [12] These provide data striping and a small number of parallel file access modes. In many cases, these access modes provide insufficient control for the application to extract good ....

Kwan, T. T., and Reed, D. A. Performance of the CM-5 Scalable File System. In Proceedings of the 1994 ACM International Conference on Supercomputing (July 1994).

No context found.

Thomas T. Kwan and Daniel A. Reed. Performance of the CM-5 Scalable File System. In Proceedings of the 8th ACM International Conference on Supercomputing, pages 156--165, Manchester, England, 1994. ACM Press.

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