E. M. Jr. Chaves, P. C. Das, T. J. Leblanc, B. D. Marsh, and M. L. Scott. Kernel-kernel communication in a sharedmemory multiprocessor. Concurrency: Practice and Experience, 5(3):171--191, May 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....a specific system [5, 6, 9, 22, 26, 28] These systems were mostly uniprocessor or small scale multiprocessor systems trying to scale up to larger systems. Other workon locality issues in operating system structure were mostly either done in the context of earlier non cachecoherent NUMA systems [8] , or, as in the case of Plan 9, were not published [25] Two projects that were aimed explicitly at large scale multiprocessors were Hive [7] and the precursor to Tornado, Hurricane [30] Both independently chose a clustered approach by connecting multiple small scale systems to form either, ....

E. M. Jr. Chaves, P. C. Das, T. J. Leblanc, B. D. Marsh, and M. L. Scott. Kernel-kernel communication in a sharedmemory multiprocessor. Concurrency: Practice and Experience, 5(3):171--191, May 1993.

....that permit the association of arbitrary service contracts with objects and object invocations. Similarly, research in high performance computing and in operating systems has derived gains in performance and reliability from the dynamic configuration of multiprocessor synchronization constructs[9, 36], of network communications[27, 26] of file systems[39] and of distributed object abstractions[12, 3] Separation and configuration can also be the means of maintaining backward compatibility with existing applications type dependent interfaces or with previous versions of operating systems, ....

E. Chaves Jr., P. Das, T. LeBlanc, B. Marsh, and M. Scott. Kernel-Kernel Communication in a Shared-Memory Multiprocessor. Concurrency: Practice and Experience, 5(3):171--192, May 1993.

....(communication overhead) For example, for the BBN Butterfly, the ratio of access costs to local vs. remote memory is approximately 1:12, which implies that the costs of executing an operation on a shared abstraction strongly depends on the number of remote references performed by the operation[6]. Since this ratio tends to be even worse on modern NUMA machines and in order to reduce contention and take advantage of locality, implementations of shared void find best tour( work queue, best tour ) tsp queue t work queue; tsp tour t best tour; f tsp node t node, left node; int i, ....

....multiprocessors: 1) it tends to improve the locality of reference of programs by removing remote references, and (2) it provides implicit synchronization for cooperating threads. The tradeoffs between remote invocation and remote memory access are empirically evaluated on the BBN Butterfly in [6, 8]. Results reported by the authors show that the overheads associated with explicit synchronization and remote references increase with increasing sizes of remotely accessed data and code (i.e. with the complexity of remote operations) whereas the overheads associated with remote invocation do ....

[Article contains additional citation context not shown here]

E.M. Chaves, Jr., P.C. Das, T.L. LeBlanc, B.D. Marsh, and M.L. Scott. Kernel-kernel communication in a shared-memory multiprocessor. Concurrency: Practice and Experience, 5(3):171--192, May 1993.

....multiprocessors, or the basic kernel model in short (Fig. 2) 3. 2 Direct Access Method and Remote Invocation Method There are two approaches to implementing an OS kernel on function distributed shared memory multiprocessors: the direct access method and the remote invocation method (Chaves et al. 1991). With the direct access method, when a task oper Local Memory I O MPU Local Memory I O MPU : local task Fig. 2. Basic Kernel Model ates on a resource on another processor, it directly accesses the control block of the resource located on the local memory of the processor. In implementing a ....

Chaves, Jr., E. M., Das, P. C., LeBlanc, T. J., Marsh, B. D., and Scott, M. L. (1991). KernelKernel Communication in a Shared-Memory Multiprocessor.

....internally in different ways, invisibly to services users. Examples of such uses are the internally parallel operating system servers offered in the Eden system [134] or in CHAOS [213, 92] and Presto [24, 23] the association of protection boundaries with certain objects as intended in Psyche [80], or the internally fragmented objects offered by Shapiro [227, 229, 98] for distributed systems, in Topologies [211] for hypercube machines, and in Distributed Shared Abstractions [57] for multiprocessor engines. Unresolved issues with object oriented operating systems include the efficient ....

.... required for certain distributed programs [103] In addition, as with remote procedure calls, it is unclear what levels of machine and language support are required for efficient implementation of object invocations (e.g. for parameter marshaling [103] or for crossing protection boundaries [258, 80]) However, object oriented operating systems are likely to become increasingly important in part (1) because of the wide range of parallel architectures (or even sequential machines ranging from digital pages to workstations) to be supported by future parallel operating systems and (2) because ....

[Article contains additional citation context not shown here]

Jr. E.M. Chaves, P.C. Das, T.L. LeBlanc, B.D. Marsh, and M.L. Scott. Kernel-kernel communication in a shared-memory multiprocessor. Concurrency: Practice and Experience, 5(3):171--192, May 1993.

.... separated from the applications basic functionalities and adjusted at runtime[20, 28, 72] Similarly, research in high performance programming and in operating systems has derived gains in performance and reliability from the dynamic configuration of multiprocessor synchronization constructs[18, 49], of network communications[34] of file systems[52] and of distributed object abstractions[10] In all of these cases, basic component functionality is separated from the implementation attributes affecting performance. Separation and configuration can also be the mechanisms for maintaining ....

J. E.M. Chaves, P. Das, T. LeBlanc, B. Marsh, and M. Scott. Kernel-kernel communication in a shared-memory multiprocessor. Concurrency: Practice and Experience, 5(3):171--192, May 1993.

....one would not want to replicate actively write shared data. Inter cluster communication is needed i) to access remote data that is not replicated, ii) to replicate and migrate data, and iii) to maintain data consistency. Three different mechanisms can be used for inter cluster communication [Chaves 1993]. First, shared memory access is always a viable option on a shared memory multiprocessor, particularly for light weight operations that make only a few references. For instance, our implementation uses shared memory for table lookup to locate objects such as page descriptors. However, we prefer ....

Chaves, E., Das, P.C., LeBlanc T.J., Marsh, B.D. and Scott, M.L. 1993. Kernel-kernel communication in a shared-memory multiprocessor. Concurrency: Practice and Experience, 5, 3, (May), pp. 171--191.

....one would not want to replicate actively write shared data. Inter cluster communication is needed i) to access remote data that is not replicated, ii) to replicate and migrate data, and iii) to maintain data consistency. Three different mechanisms can be used for intercluster communication [13]. First, shared memory access is always a viable option on a shared memory multiprocessor, particularly for light weight operations that make only a few references. For instance, our implementation uses shared memory for table lookup to locate objects such as page descriptors. However, we prefer ....

E. Chaves, T. J. LeBlanc, B. D. Marsh, and M. L. Scott. Kernel-kernel communication in a sharedmemory multiprocessor. In Proc. Second Symposium on Distributed and Multiprocessor Systems, pages 105--116, Atlanta, Georgia, March 1991. Usenix.

....lock structuring issues, and hence reduces code complexity, which can lead to improved performance and scalability. For example, Chaves reports that the fine grained locking used in an unclustered system significantly increases the length of the critical path, even when there is no lock contention [7]. As well, deadlock can be a problem when several fine grained locks must be held simultaneously. Because contention for a lock is primarily limited to the number of processors in a cluster, clustering allows for coarser grained locking, as we show in Section 5. Using clusters to structure an ....

E. Chaves, T. J. LeBlanc, B. D. Marsh, and M. L. Scott. Kernel-kernel communication in a shared-memory multiprocessor. In Proc. Second Symposium on Distributed and Multiprocessor Systems, pages 105--116, Atlanta, Georgia, March 1991. Usenix.

....view of locking, hierarchical clustering provides two major benefits. First, it bounds contention for both coarse grained locks and reserve bits, since RPCs are used instead of shared memory to access remote entries (Chaves et al. discuss some of the trade offs between using RPC and shared memory [3]) Second, it increases lock bandwidth by i) instantiating per cluster system data structures (such as the hash table in Figure 2) each with its own coarse grained lock, and by ii) replicating read shared objects (such as X in Figure 2) each with its own reserve bit. Although hierarchical ....

E. Chaves, P.C Das, T. J. LeBlanc, B. D. Marsh, and M. L. Scott. Kernel-kernel communication in a shared- memory multiprocessor. Concurrency: Practice and Experience, 5(3):171--191, May 1993.

No context found.

E.M. Chaves Jr., P.C. Das, T.L. LeBlanc, B.D. Marsh, and M.L. Scott. Kernel-kernel communication in a shared-memory multiprocessor. Concurrency: Practice and Experience, 5(3):171--192, May 1993.

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