Performance evaluation of the Orca shared-object system

Bal, Henri E.; Bhoedjang, Raoul; Hofman, Rutger F. H.; Jacobs, Ceriel J. H.; Langendoen, Koen; Rühl, Tim; Kaashoek, M. Frans

doi:10.1145/273011.273014

Cited by 100 publications

(71 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Satin's work stealing algorithms are implemented on top of the Panda communication library [2], which has efficient implementations on a variety of networks. On the Myrinet network (which we use for the measurements in this paper), Panda is implemented on top of the LFC [4] network interface protocol.…”

Section: The Satin Systemmentioning

confidence: 99%

Efficient load balancing for wide-area divide-and-conquer applications

2001

Self Cite

View full text Add to dashboard Cite

Divide-and-conquer programs are easily parallelized by letting the programmer annotate potential parallelism in the form of spawn and sync constructs. To achieve efficient program execution, the generated work load has to be balanced evenly among the available CPUs. For single cluster systems, Random Stealing (RS) is known to achieve optimal load balancing. However, RS is inefficient when applied to hierarchical wide-area systems where multiple clusters are connected via wide-area networks (WANs) with high latency and low bandwidth.In this paper, we experimentally compare RS with existing loadbalancing strategies that are believed to be efficient for multi-cluster systems, Random Pushing and two variants of Hierarchical Stealing. We demonstrate that, in practice, they obtain less than optimal results. We introduce a novel load-balancing algorithm, Clusteraware Random Stealing (CRS) which is highly efficient and easy to implement. CRS adapts itself to network conditions and job granularities, and does not require manually-tuned parameters. Although CRS sends more data across the WANs, it is faster than its competitors for 11 out of 12 test applications with various WAN configurations. It has at most 4% overhead in run time compared to RS on a single, large cluster, even with high wide-area latencies and low wide-area bandwidths. These strong results suggest that divideand-conquer parallelism is a useful model for writing distributed supercomputing applications on hierarchical wide-area systems.

show abstract

Section: The Satin Systemmentioning

confidence: 99%

Efficient load balancing for wide-area divide-and-conquer applications

2001

Self Cite

View full text Add to dashboard Cite

show abstract

“…Orca [4,3,2] is an object based programming language and distributed shared memory system (DSM). Orca is based on distributed coherent objects, e.g.…”

Section: Related Workmentioning

confidence: 99%

DOLCLAN – Middleware Support for Peer-to-Peer Distributed Shared Objects

Bardram

Mogensen

2007

Distributed Applications and Interoperable Systems

View full text Add to dashboard Cite

Abstract. Contemporary object-oriented programming seeks to enable distributed computing by accessing remote objects using blocking remote procedure calls. This technique, however, suffers from several drawbacks because it relies on the assumption of stable network connections and synchronous method invocations. In this paper we present an approach to support distributed programming, which rely on local object replicas keeping themselves synchronized using an underlying peer-to-peer infrastructure. We have termed our approach Peer-to-peer Distributed Shared Objects (PDSO). This PDSO approach has been implemented in the DOLCLAN framework. An evaluation demonstrates that DOLCLAN can be utilized to create a real distributed collaborative system for ad-hoc collaboration in hospitals, which demonstrates that the approach can support the creation of non-trivial distributed applications for pervasive computing.

show abstract

“…Orca [5] and Jade [31] are object-oriented parallel programming languages. With distributed JVM as the executing platform, Java can also be considered a parallel programming language.…”

Section: Related Workmentioning

confidence: 99%

“…It may however incur heavy diff overhead compared with conventional single-writer protocol. Another choice is between the update protocol (e.g., Orca [5]) and the invalidate protocol in many page-based DSM systems (e.g., [20] [9]). The update protocol can do prefetching to make the data available before the access, but it may send much unneeded data compared with the invalidate protocol.…”

Section: Introductionmentioning

confidence: 99%

On the design of global object space for efficient multi-threading Java computing on clusters

Fang¹,

Wang²,

Lau³

2003

Parallel Computing

View full text Add to dashboard Cite

A distributed Java Virtual Machine supports transparent parallel execution of multi-threaded Java programs on a cluster of computers. It provides an alternative platform for high performance scientific computations. In this paper, we present the design of a global object space in distributed JVM, which virtualizes a single Java object heap across machine boundaries to facilitate transparent object accesses. We leverage runtime object connectivity information to detect objects, called distributed-shared objects, that are reachable from threads at different nodes to address memory consistency issues in the distributed JVM. With the detection of distributed-shared objects, we can reduce memory consistency maintenance overhead and provide efficient distributed garbage collection.We propose a framework to characterize object access patterns, along three orthogonal dimensions. Under the framework, we propose and apply three effective adaptations to the cache coherence protocol to optimize several access patterns that frequently appear during application execution, including an object home migration method that adapts to the single-writer access pattern, synchronized method migration that allows the remote execution of a synchronized method at the home node of its locked object, and object pushing that uses the object connectivity information to adapt to the producer-consumer access pattern.

show abstract

Performance evaluation of the Orca shared-object system

Cited by 100 publications

References 34 publications

Efficient load balancing for wide-area divide-and-conquer applications

Efficient load balancing for wide-area divide-and-conquer applications

DOLCLAN – Middleware Support for Peer-to-Peer Distributed Shared Objects

On the design of global object space for efficient multi-threading Java computing on clusters

Contact Info

Product

Resources

About