R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, October 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... Peacock et al. 28] 29] kernel memory allocation facilities developed by McKenny et al. 25] fair fast scalable reader writer locks developed by Krieger et al. 16] performance measurement kernel device driver developed by Anderson et al. 1] and the intra node data structures used by Stets et al.[35]. The traditional approach of adding locks and selectively redesigning also does not explicitly lead to increased locality. Chapin et al. studied the memory system performance of a commercial Unix system, parallelized to run e#ciently on the 64 processor large Stanford DASH multiprocessor[6] ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and Michael Scott. Cashmere-2L: Software coherent shared memory on a clustered remote-write 99 network. In Proc. of the 16th ACM Symp. on Operating Systems Principles (SOSP-16), October 1997.

....software DSM without the need for complex relaxed consistency models. A simple technique like page freezing is sufficient to obtain the same performance level with a strict consistency model. 5. Related Work Since IVY [9] a lot of work has been done on DSM systems. The Cashmere 2L system [14] is implemented on a cluster of SMP nodes and uses the write only based Memory Channel. Memory coherency is maintained in hardware within each SMP node and uses a home based Lazy Release Consistency model between nodes. Cashmere 2L is based, like TreadMarks [8] on the twin diff technique that ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy and M. Scott. Cashmere-2L : Software coherent shared memory on a clustered remote-write network. 16 SOSP. pp. 170-183, October 1997.

....libraries have been upgraded to work with CLUMPs [3] 4] 2] The performance results presented in [4] claim that a cluster of uniprocessors can be faster than a CLUMP. Shared virtual memory environments (DSVM) provide another alternative to unify the memory models, as presented in [5] [6] [7] 8] Performance of the DSVM model mainly relies on the efficiency of the memory consistency protocol that is used. Two other approaches can be noticed. Split C [9] is a parallel extension to C that supports access to a global address space on distributed memory architectures. Recently, ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and Michael Scott. Cashmere-2L: Software coherent shared memory on a clustered remote-write network. In Proc. of the 16th ACM Syrup. on Operating Systems Principles (SOSP-16), Oc- tober 1997.

....are also presented for the integer sort, FFT and searching problems. The shared Virtual Memory environments provide opportunity to program the CLUMP with the shared memory model. Several projects have published some design and performance results about the CLUMP:J7] Shasta [8] Cashmere 2L [9] and SoftFLASH [10] As for the monoprocessor platforms, the performance of DVSM mainly relies on the protocol efficiency. Other programming paradigms, initially developed in the context of the single processor node parallel computers, may also unify the memory model for the CLUMP. Split C [11] ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and Michael Scott. Cashmere-2L: Software coherent shared memory on a clustered remote-write network. In Proc. of the 16th ACM Syrup. o Operating Systems Principles (SOSP-16), October 1997.

....and collective communications. 1.2 Shared Memory SMM Shared Virtual Memory environments provides the opportunity to program the CLUMP with the shared memory model. Several projects have already published some design and performance results about the CLUMP:J7] Shasta [8] Cashmere 2L [9] and SoftFLASH [10] As for the mono processor platforms, the performance of DVSM mainly relies on the protocol efficiency. Other programming paradigms, initially developed in the context of the single processor node parallel computers, may also unify the memory model for the CLUMP. Split C [11] ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and Michael Scott. Cashmere-2L: Software coherent shared memory on a clustered remote-write network. In Proc. of the 16th ACM Symp. on Operating Systems Principles (SOSP-16), October 1997.

....a SDSM system is to reduce communication overhead between processors. This is usually at the cost of increased computation and or loss of programming abstraction. This paper presents a comparison of four software distributed shared memory systems, namely TreadMarks[1] OpenMP SMP[6] Cashmere 2L[15] and Shasta[14] We compare and contrast them within the context of a taxonomy which we feel is appropriate to model software distributed shared memory systems. We present this taxonomy after a brief de nition of a SDSM. This is followed by descriptions of the four systems. We close the paper with ....

....which aid in the compilation and execution of the program on a speci c architecture. 3. 5 Support for SMPs The availability and relatively low cost of symmetric multiprocessors (SMPs) has resulted in many SDSM system designers to expand their systems to support networks of these machines[6, 12, 15, 11]. By running their system on networks of SMPs, they are able to exploit the hardware cache coherence between processors within an SMP node, thereby reducing the overall communication overhead that is required to maintain coherence across the system as a whole. 4 4 SDSM Systems In this section, ....

[Article contains additional citation context not shown here]

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, 10 and M. L. Scott. Cashmere-2L: Software coherent shared memory on a clustered remote-write network. In Sixteenth ACM Symposium on Operating Systems Principles, October 1997. 11

....diffs, quickly applies diffs, and minimizes the space used to store and transmit diffs. We show that the algorithm exhibits significantly reduced memory consumption and network costs when compared to existing differencing mechanisms. The basic method employed by all existing diff mechanisms, [5, 13, 18, 19] can be summarized as follows. Initially write access is disabled for the shared data region (we will refer to them as shared blocks) When a process attempts to modify the block, a write access fault occurs and invokes an interrupt handler. The handler creates an immutable copy of the block, ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2l: Software coherent shared memory on a clustered remotewrite network. In Proc. SOSP '97, October 1997.

....put and get functionality is available. However, no global virtual memory system like the SCI VM presented above is provided. Other well known work for NUMA systems in academia can be found at Princeton University in the Shrimp project [3] and at the University of Rochester in the Cashmere project [21]. 6 Conclusions and Future work In this paper we have shown that a cluster of PCs connected with the help of the SCI interconnection technology can efficiently support both major programming models for parallel systems, message passing and shared memory. The message passing capabilities have ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. CASHMERE-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proceedings of SOSP'97, Oct. 1997.

....hierarchy and especially the population of the memory banks have a signi cant impact on performance. Other programming models have been designed for the CLUMPs. For example, Shared virtual memory environments (DSVM) provide another alternative to unify the memory models, as presented in [7] [8] [9] 10] Recently, OpenMP [11] has been implemented on a cluster of SMPs on top of the Treadmark DSM system. In[12] a hybrid shared memory distributed memory model is proposed for CLUMPs. Communications between nodes use message passing and remote memory operations. The model is based on the ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and Michael Scott. Cashmere-2L: Software coherent shared memory on a clustered remote-write network. In Proc. of the 16th ACM Symp. on Operating Systems Principles (SOSP-16), October 1997.

....model (HMM) In the SMM approach, a mechanism is provided to unify the memory model that the programmer sees. Some message passing libraries have been updated to provide CLUMP with a single memory model [4] 1] 3] Using shared virtual memory environments (DSVM) is another approach for SMM [5] [6] [7] 8] Recently, OpenMP has been implemented on a cluster of SMPs on top of the Treadmark DSM system [9] When using share virtual memory, performance mainly relies on the efficiency of the memory consistency protocol. The HMM approach, previously used for Parallel Vector Processors, has been ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and Michael Scott. Cashmere-2L: Software coherent shared memory on a clustered remote-write network. In Proc. of the 16th ACM Symp. on Operating Systems Principles (SOSP-16),October 1997.

....targeting both scientific and commercial applications. Shared memory clusters are an attractive approach to providing a#ordable and scalable compute cycles and I O. For this reason, there has recently been a lot of work on designing e#cient shared virtual memory (SVM) protocols for such clusters [23,16,26,13]. These protocols take advantage of features provided by SANs, such as low latencies for short messages and direct remote memory operations with no remote processor intervention [12,10,9] to improve system performance and scalability [16] Providing a shared memory programming abstraction on ....

.... improve system performance and scalability [16] Providing a shared memory programming abstraction on clusters has made it easier to run applications that have been written for more traditional, tightly coupled multiprocessors (both shared bus and distributed shared memory machines) Recent work [23,16,26] has shown that the performance of SVM clusters is competitive for wide ranges of applications to more traditional, tightly coupled multiprocessors. For instance, the authors in [16] find that a 64 processor cluster o#ers, for most SPLASH 2 [24] applications (after a number of optimizations) at ....

[Article contains additional citation context not shown here]

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L.Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles (SOSP-16), Oct. 1997.

....todays state of the art SANs. The focus is on protocol level issues for extending the Cashmere protocol to clusters of SMPs and interconnection network with direct remote memory access capabilities. A subset of the features examined in the simulation studies was implemented on an actual cluster [14], which, however, corresponds more to state of the art clusters that can be built with today s technology. The SoftFLASH system [4] provided a sequentially consistent software shared memory layer on top of 8 processor SMP nodes. They find that the cost for page invalidations within each node is ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles (SOSP-16), Oct. 1997.

No context found.

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles, St. Malo, France, Oct. 1997.

....of Section 1. The InterWeave API used in the example is explained in more detail in the following sections. For consistency with the example we present the C version of the API. Similar versions exist for C , Java, and Fortran. InterWeave is designed to interoperate with our Cashmere S DSM system [32]. Together, these systems integrate hardware coherence and consistency within multiprocessors (level 1 sharing) S DSM within tightly coupled clusters (level 2 sharing) and version based coherence and consistency across the Internet (level 3 sharing) At level 3, InterWeave uses ....

....advantage of InterWeave s ability to share data across heterogeneous platforms. Astroow is a computational uid dynamics system used to study the birth and death of stars. The simulation engine is written in Fortran, and runs on a cluster of four AlphaServer 4100 5 600 nodes under the Cashmere [32] S DSM system. As originally implemented it dumps its results to a le, which is subsequently read by a visualization tool written in Java and running on a Pentium desktop. We used InterWeave to connect the simulator and visualization tool directly, to support on line visualization and steering. ....

[Article contains additional citation context not shown here]

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles, St. Malo, France, Oct. 1997.

....modes; and (c) reduce the number of trivial invocations used simply to put or get data. Unfortunately, sharing is significantly more complex in a heterogeneous, wide area network environment than it is in software distributed shared memory (S DSM) systems such as TreadMarks [1] and Cashmere [14]. With rare exceptions, S DSM systems assume that clients are part of a single program, written in a single language, running on identical hardware nodes on a system area network. InterWeave must support coherent, persistent sharing among programs written in multiple languages, running on multiple ....

....advantage of InterWeave s ability to share data across heterogeneous platforms. Astroflow is a computational fluid dynamics system used to study the birth and death of stars. The simulation engine is written in Fortran, and runs on a cluster of four AlphaServer 4100 5 600 nodes under the Cashmere [14] SDSM system. As originally implemented, it dumps its results to a file, which is subsequently read by a visualization tool written in Java and running on a Pentium desktop. We used InterWeave to connect the simulator and visualization tool directly, to support on line visualization and steering. ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered RemoteWrite Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles, pages 170--183, St. Malo, France, Oct. 1997.

....allows processes to share arbitrarily complex structured data safely and efficiently, with ordinary reads and writes. Many researchers (ourselves among them) have developed software distributed shared memory (S DSM) systems to extend this programming model into message based environments [1, 17, 21]. Object based systems can of course be implemented on top of shared memory, but the lower level interface suffices for many applications. Unfortunately, despite some 15 years of research, S DSM remains for the most part a laboratory curiosity. The explanation, we believe, lies with the choice of ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered RemoteWrite Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles, pages 170--183, St. Malo, France, Oct. 1997.

....typically attempt to increase the locality of shared state through ad hoc, application specific caching or replication protocols built on top of RPC based systems such as Sun RPC, Java RMI, CORBA, and .NET. Software distributed shared memory (S DSM) systems such as TreadMarks [3] and Cashmere [39] automate the management of shared state for local area clusters, but they do not extend in any obvious way to geographic distribution. As a general rule, they assume that sharing processes are part of a single parallel program, running on homogeneous hardware, with communication latencies and ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. L. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles, pages 170-- 183, St. Malo, France, Oct. 1997.

....this demand, we are developing a system, known as InterWeave [8] that allows the programmer to map shared data into program components regardless of location or machine type, and to transparently access that data once mapped. InterWeave represents a merger and extension of our previous Cashmere [19, 20] and InterAct [15] projects. Once shared data has been mapped, InterWeave can support hardware coherence and consistency within multiprocessors (level 1 sharing) Cashmere style software distributed shared memory (S DSM) within tightly coupled clusters (level 2 sharing) and InterAct style ....

....tightly coupled) levels. At levels 1 and 2, InterWeave inherits Cashmere s integration of intra SMP hardware cache coherence with cluster level VM based lazy release consistency. In particular, it employs two way diffing to avoid the need for TLB shootdown when processes synchronize across nodes [19], and relies on low latency user level messages for efficient synchronization, directory management, and write notice propagation [20] In a similar vein, consistency at level 3 employs the twins, diffs, write notices, and home node copies already maintained at level 2. At the third level, data ....

[Article contains additional citation context not shown here]

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered RemoteWrite Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles, St. Malo, France, Oct. 1997.

....actually needed, and only when any already cached copies are not recent enough to use. Unfortunately, sharing is significantly more complex in a heterogeneous, wide area network (WAN) environment than it is in software distributed shared memory (S DSM) systems such as TreadMarks [1] and Cashmere [19]. With rare exceptions, S DSM systems assume that clients are part of a single program, written in a single language, running on identical hardware nodes on a system area network. InterWeave must support coherent, persistent sharing among programs written in multiple languages, running on multiple ....

....of InterWeave s ability to share data across heterogeneous platforms. Astroflow [8] is a computational fluid dynamics system used to study the birth and death of stars. The simulation engine is written in Fortran, and runs on a cluster of four AlphaServer 4100 5 600 nodes under the Cashmere [19] S DSM system. As originally implemented, it dumps its results to a file, which is subsequently read by a visualization tool written in Java and running on a Pentium desktop. We used InterWeave to connect the simulator and visualization tool directly, to support on line visualization and steering. ....

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proc. of the 16th ACM Symp. on Operating Systems Priciples, St. Malo, France, Oct. 1997.

....of Section 1. The InterWeave API used in the example is explained in more detail in the following sections. For consistency with the example we present the C version of the API. Similar versions exist for C , Java, and Fortran. InterWeave is designed to interoperate with our Cashmere S DSM system [32]. Together, these systems integrate hardware coherence and consistency within multiprocessors (level 1 sharing) S DSM within tightly coupled clusters (level 2 sharing) and version based coherence and consistency across the Internet (level 3 sharing) At level 3, InterWeave uses ....

....advantage of InterWeave s ability to share data across heterogeneous platforms. Astroflow is a computational fluid dynamics system used to study the birth and death of stars. The simulation engine is written in Fortran, and runs on a cluster of four AlphaServer 4100 5 600 nodes under the Cashmere [32] S DSM system. As originally implemented it dumps its results to a file, which is subsequently read by a visualization tool written in Java and running on a Pentium desktop. We used InterWeave to connect the simulator and visualization tool directly, to support on line visualization and steering. ....

[Article contains additional citation context not shown here]

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proc. of the 16th ACM Symp. on Operating Systems Principles, St. Malo, France, Oct. 1997.

No context found.

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2L: Software Coherent Shared Memory on a Clustered Remote-Write Network. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, October 1997.

No context found.

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2l: Software coherent shared memory on a clustered remote-write network. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, October 1997.

No context found.

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and M. Scott. Cashmere-2l: Software coherent shared memory on a clustered remote write network. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, pages 170--183, Oct 1997.

No context found.

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, H. Kontothannassis, Parhasarathy, and M. Scott. Cashmere-2L: software coherent shared memory on a clustered remotewrite network. In Proceedings of the 16th ACM Symposium on Operating System Principles, pages 170--183, Oct. 1997.

No context found.

R. Stets, S. Dwarkadas, N. Hardavellas, G. Hunt, L. Kontothanassis, S. Parthasarathy, and Michael Scott. Cashmere-2L: Software coherent shared memory on a clustered remote-write network. In Proc. of the 16th ACM Syrup. o Operating Systems Principles (5'0SP-16), October 1997.

First 50 documents  Next 50

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