| D. Kotz and T. Cai, "Exploring the uses of I/O nodes for computation in a MIMD multiprocessor, " Technical Report TR 95-253, Dartmouth College, 1995. |
....each computing node. This would unfortunately require having at least two, three or four disks per computing node. Another solution would be to organize the cluster nodes themselves into one of more disk striping units implementing the same 2 architecture as a RAID but without a shared controller [2, 5]. The two major advantages of the solution are its low space overhead and its very good reliability. Organizing the cluster nodes into striping units would however destroy the locality of disk accesses and consequently increase the network traffic among the nodes. W N N S E E 25 S 25 ....
....will have to allocate disk space for them. We do not believe that the processing overhead is a major problem because the likelihood of finding spare processor cycles is bound to increase thanks to the ever increasing speeds of processors. This observation was recently confirmed by Kotz and Cai [5]. The space overhead issue could be more critical but there are at least two potential approaches to reduce it. Consider the case of a computing node of figure 2, which has four neighbors, respectively named N, E, S and W. Let us assume for the sake of simplicity that its backup areas for the ....
D. Kotz and T. Cai, "Exploring the uses of I/O nodes for computation in a MIMD multiprocessor, " Technical Report TR 95-253, Dartmouth College, 1995.
....on each computing node. This would unfortunately require having at least two, three or four disks per computing node. Another solution would be to organize the cluster nodes themselves into one of more disk striping units implementing the same architecture as a RAID but without a shared controller [2, 5]. The two major advantages of the solution are its low space overhead and its very good reliability. Organizing the cluster nodes into striping units would however destroy the locality of disk accesses and consequently increase the network traffic among the nodes. The solution we propose offers ....
....will have to allocate disk space for them. We do not believe that the processing overhead is a major problem because the likelihood of finding spare processor cycles is bound to increase thanks to the ever increasing speeds of processors. This observation was recently confirmed by Kotz and Cai [5]. The space overhead issue could be more critical but there are at least two potential approaches to reduce it. 3 W N N S E E 25 S 25 25 25 Figure 2: The same node receiving copies of the critical data of its four neighbors Consider the case of a computing node of figure 2, which ....
D. Kotz and T. Cai, "Exploring the Uses of I/O Nodes for Computation in a MIMD Multiprocessor, " Technical Report TR 95-253, Dartmouth College, 1995.
....for specific applications and or specific architectures. As an example, take the area of persistent storage for parallel applications. Different parallel file systems have been elaborated specifically for scientific applications[13] 8] Many are designed for machines that have dedicated I O nodes [18, 8]. Some offer specialized interfaces to support multidimensional matrices [26] or interfaces to support joint I O requests from many processes [4] Good results are achieved, but the efforts deployed can not easily be used for other applications or architectures. This paper describes an I O ....
D. Kotz, Ting Cai. "Exploring the use of I/O nodes for computations in a MIMD multiprocessor." In Workshop for I/O on Parallel and Distributed Systems.
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