this paper describes a general technique to identify control flow errors in parallel programs, which can be automated into a compiler. the compiler builds a system of linear equations that describes the global control...
详细信息
this paper describes a general technique to identify control flow errors in parallel programs, which can be automated into a compiler. the compiler builds a system of linear equations that describes the global control flow of the whole program. Solving these equations using standard techniques of linear algebra can locate a wide range of control flow bugs at compile time. this paper also describes an implementation of this control flow analysis technique in a prototype compiler for a well-known parallel programming language. In contrast to previous research in automated parallel program analysis, our technique is efficient for large programs, and does not limit the range of language features.
this paper presents Pangaea, an inter-datacenter key-value store that keeps reasonable expandability of storage capacity, data lookup latency and data transfer speed. Pangaea uses two techniques called multi-layered D...
详细信息
ISBN:
(纸本)9781479920815
this paper presents Pangaea, an inter-datacenter key-value store that keeps reasonable expandability of storage capacity, data lookup latency and data transfer speed. Pangaea uses two techniques called multi-layered DHT (ML-DHT) and local-first data rebuilding (LDR). ML-DHT provides a global and consistent index of key-value pairs with efficient routings in inter-datacenter environments. LDR reduces inter-datacenter data transfer by using erasure coding techniques.
Dynamic distributedsystems allow processes to join and leave the system, so the number of processes participating in a computation varies over time. Examples of dynamic distributedsystems include peer-to-peer networ...
详细信息
ISBN:
(纸本)9781479920815
Dynamic distributedsystems allow processes to join and leave the system, so the number of processes participating in a computation varies over time. Examples of dynamic distributedsystems include peer-to-peer networks, sensor networks, mobile ad-hoc networks, and many more. A fundamental problem in any distributed system is to find consensus among the processes on a common input value. For dynamic distributedsystems, it is not entirely clear how the problem should be formulated, as processes can join and leave before consensus is reached. We formulate and solve a strong version of the consensus problem in dynamic distributedsystems in the presence of Byzantine faulty processes. We show that one cannot improve upon our algorithm in terms of the bound on the number of processes. For stochastic dynamic distributedsystems, we determine the probability that a set of processes can reach strong consensus.
As information systems develop into larger and more complex implementations, the need for survivability increases. When components are exported from a remote system to a local system under different administrative set...
详细信息
ISBN:
(纸本)9780889867741
As information systems develop into larger and more complex implementations, the need for survivability increases. When components are exported from a remote system to a local system under different administrative settings and deployed in different environments we have to guarantee the proper execution of those remote components in the currently working environment because the remote components may have failures or malicious codes that can affect the local computing environment. In this paper, we address a multiple-aspect testing approach, which is able to increase the possibility of detecting a fault in a component by having various ways possible to conduct the test on the components. We have implemented a prototype that provides for the multiple-aspect testing for the downloaded components in runtime without access to the source code. Finally, the implementation of the multiple-aspect testing is evaluated and analyzed in terms of detection performance and accuracy.
To achieve good parallel efficiency, applications using structured adaptive mesh refinement (SAMR) need to repeatedly repartition and redistribute the underlying dynamic grid hierarchy. However, no single partitioner ...
详细信息
ISBN:
(纸本)9780889867741
To achieve good parallel efficiency, applications using structured adaptive mesh refinement (SAMR) need to repeatedly repartition and redistribute the underlying dynamic grid hierarchy. However, no single partitioner works well for all application and computer states. this paper presents the implementation and evaluation of a patch-based partitioner for SAMR grid hierarchies. the partitioner results in a good and stable load balance, on average 3.1%. Space-filling curves are used to reduce the high communication volumes that are inherent in this type of partitioner. the partitioner will be a part of the Meta-Partitioner, a partitioning framework that automatically selects, configures and invokes good-performing partitioners for general SAMR applications. Access to a large number of complementing partitioners is essential for the Meta-Partitioner. the presented partitioner will help to significantly decrease run-times for SAMR applications where load balance is the main priority.
this paper introduces invasive computing, a new paradigm for programming parallel architectures. the goals are to enable the development and execution of resource aware programs that can dynamically allocate and free ...
详细信息
ISBN:
(纸本)9780889868649
this paper introduces invasive computing, a new paradigm for programming parallel architectures. the goals are to enable the development and execution of resource aware programs that can dynamically allocate and free new resources in phases with more parallelism. To allocate more resources, applications use the invade operation and to free them the retreat. the research is conducted within the framework of the Transregional Collaborative Research Centre 89 funded by the German Science Foundation.
We present a novel dynamic on-the-fly race detection mechanism called parallel Nondeterminator to check for determinacy races during the parallel execution of a program with Spawn-Sync parallelism. the parallel Nondet...
详细信息
We present a novel dynamic on-the-fly race detection mechanism called parallel Nondeterminator to check for determinacy races during the parallel execution of a program with Spawn-Sync parallelism. the parallel Nondeterminator provides provable correctness and efficiency. Let D denote the maximum depth of the recursion in the parallel program. the worst case slowdown in execution incurred for each spawn operation is O(D), the overhead for each sync operation is O(1) and the time required to monitor any shared memory access is O(log D). Moreover, we have implemented the parallel Nondeterminator in Cilk, a parallel language developed at MIT. Boththeoretical and experimental results give strong evidences for the efficiency of our algorithm.
We propose a method for the parallel execution of applications that process continuous streams of data. Unlike pipeline-based approaches, which are frequently employed to parallelize software for multi-core processors...
详细信息
ISBN:
(纸本)9780769543284
We propose a method for the parallel execution of applications that process continuous streams of data. Unlike pipeline-based approaches, which are frequently employed to parallelize software for multi-core processors, our method supports nonlinear structures that may contain conditionals. Nonlinear structures reduce the latency for processing an element from a stream, which is particularly important for embedded systemsthat are subject to real-time constraints.
Use of complex image analysis and globally optimal techniques make the current Multiple Object Tracking (MOT) methods for video analysis computationally slow. An important issue in this context is meeting the specific...
详细信息
ISBN:
(纸本)9781450363723
Use of complex image analysis and globally optimal techniques make the current Multiple Object Tracking (MOT) methods for video analysis computationally slow. An important issue in this context is meeting the specific latency requirement for a given application while processing large scale video data. this is especially important in emergency situations such as accidents, natural calamities, and terrorist attacks. this paper introduces a latency reducing MapReduce/Hadoop-based parallel solution for MOT. the system includes an Auto-Resource Provisioning technique for determining the number of Hadoop nodes required to process the MOT job within a user specified deadline. the estimated number of nodes are then provisioned by the system and the MOT application is executed on the Hadoop cluster comprising the desired number of nodes. A prototype is built using the AWS EC2 cloud. A performance analysis is performed using measurements made on the prototype and insights gained into system behavior and performance are presented.
暂无评论