R. Bryant, 11ung-Yang Chang, B. Rosenburg, "Experience Developing the RP3 Operating System", Computing Systems, The Journal of USENIX Association, Vol. 4, No. 3, pp. 183-216, 1991  Home/Search   Document Not in Database   Summary   Related Articles   Check

....system developed for SUPRENUM [3] PEACE started in 1986 as an object based operating system relying on the microkernel approach. Lessons learned from this development led to the conclusion that a microkernel based system organization is not lightweight enough for parallel architectures ( 7] and [1]) The more promising approach is to combine the idea of program families with object orientation [2] and, thus, to abandon the microkernel approach. Since 1990 PEACE therefore runs through a metamorphosis, from an object based to an object oriented system. The paper presents a novel object model ....

R. Bryant, 11ung-Yang Chang, B. Rosenburg, "Experience Developing the RP3 Operating System", Computing Systems, The Journal of USENIX Association, Vol. 4, No. 3, pp. 183-216, 1991

....by busy waiting to coordinate the computation. If the application is executed on a different number of PEs, it will partition the work accordingly and perform flawlessly. But if a PE is reclaimed by the system during execution, the others will spin forever at the next synchronization point [84]. The assumption that the number of PEs does not change may even be embedded in the programming environment. For example, the HPF programming model assumes that the number of PEs is not necessarily known at compilation, but is fixed throughout any given execution [378] A set of intrinsic ....

....that threads that are spawned together actually execute in parallel, and therefore busy waiting can be used to synchronize them. If gang scheduling is not used, this can have grave performance implications [198, 631] Moreover, if the scheduling is not preemptive at all, it could lead to deadlock [84]. A related issue is the granularity of the interactions. In general, granularity refers to amount of computation performed between successive interactions [336] Thus in fine grain interactions, interactions (such as synchronization or transfer of data) are numerous and frequent, whereas ....

R. Bryant, H-Y. Chang, and B. Rosenburg, "Experience developing the RP3 operating system ". Computing Systems 4(3), pp. 183--216, Summer 1991.

....by busy waiting to coordinate the computation. If the application is executed on a different number of PEs, it will partition the work accordingly and perform flawlessly. But if a PE is reclaimed by the system during execution, the others will spin forever at the next synchronization point [46]. The assumption that the number of PEs does not change may even be embedded in the programming environment. For example, the HPF programming model assumes that the number of PEs is not necessarily known at compilation, but is fixed throughout any given execution [227] A set of intrinsic ....

R. Bryant, H-Y. Chang, and B. Rosenburg, "Experience developing the RP3 operating system ". Computing Systems 4(3), pp. 183--216, Summer 1991.

.... does not help with the efficient operation of a database machine 4 Operating systemlevel interference with Fortran computations may reduce the speed of the computation by reducing control of the Fortran program over the scheduling policies and hardware resources of the machine. For example, Bryant et al. (1991) described their experience with Fortran users of the IBM RP3 system. These users regarded the operating system as an adversary bent on denying them direct access to the hardware . A major problem with regard to performance of applications is that of data locality: an executing thread of control ....

Bryant, R., H. Chang and B. Rosenburg (1991) Experience developing the RP3 operating system. Usenix Association Distributed and Multiprocessor Systems (Summer), 1-18.

....is in progress since 1990. Lessons learned from the Suprenum development led to the conclusion that a microkernel based system organization is not lightweight enough for parallel architectures [14] Similar observations have been made with porting both Mach onto a sharedmemory parallel computer [2] and Chorus onto a high performance Risc architecture [16] Overcoming the performance bottleneck problem in distributed memory architectures forbids a single microkernel implementation aimed at supporting both a family of parallel operating systems and parallel applications. Rather, the operating ....

R. Bryant, Hung-Yang Chang, B. Rosenburg, "Experience Developing the RP3 Operating System", Computing Systems, The Journal of USENIX Association, Vol. 4, No. 3, pp. 183-216, 1991

....system developed for Suprenum [3] Peace started in 1986 as an objectbased operating system relying on the microkernel approach. Lessons learned from this development led to the conclusion that a microkernel based system organization is not lightweight enough for parallel architectures ( 7] and [1]) The more promising approach is to combine the idea of program families with object orientation [2] and, thus, to abandon the microkernel approach. Since 1990 Peace therefore runs through a metamorphosis, from an object based to an object oriented system. The paper presents a novel object model ....

R. Bryant, Hung-Yang Chang, B. Rosenburg, "Experience Developing the RP3 Operating System", Computing Systems, The Journal of USENIX Association, Vol. 4, No. 3, pp. 183-216, 1991

....are to be modified to maximize per processor locality, how are these modifications to be achieved One might simply leave it up to the application programmer, but this is unlikely to be acceptable. Experience with shared memory systems of the past (e.g. the BBN Butterfly [17] and the IBM RP3 [6]) suggests that achieving enough locality to obtain nearlinear speedups on large numbers of processors is a very difficult task, and the growing disparity between processor and memory speeds suggests that the difficulty will increase in future years [22] For many of the most demanding parallel ....

R. Bryant, H.-Y. Chang and B. Rosenburg, "Experience Developing the RP3 Operating System," Proceedings of the Second USENIX Symposium on Experiences with Distributed and Multiprocessor Systems, 21-22 March 1991, pp. 1-18.

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