The following topics are dealt with: NLP (natural language processing) technologies for MLIM (multi-lingual information management); system issues in MLIM; and multilingual text processing.
The following topics are dealt with: NLP (natural language processing) technologies for MLIM (multi-lingual information management); system issues in MLIM; and multilingual text processing.
CCL (Checkpointing and Communication Library) is a recently developed software in support of optimistic parallelsimulation on myrinet based clusters. Beyond classical low latency message delivery functionalities, thi...
详细信息
ISBN:
(纸本)0769519709
CCL (Checkpointing and Communication Library) is a recently developed software in support of optimistic parallelsimulation on myrinet based clusters. Beyond classical low latency message delivery functionalities, this library implements CPU offloaded, semi-asynchronous checkpointing functionalities based on data transfer capabilities provided by a programmable DMA engine on board of myrinet network cards. The latest version of CCL (v2.4), designed for M2M-PCI32C myrinet cards, only supports monoprogrammed semi-asynchronous checkpoints. This forces re-synchronization between CPU and DMA activities each time a new checkpoint request must be issued at the simulation application level while the last issued one is still being carried out by the DMA engine. In this paper we present CCL v3.0 that, exploiting hardware features of more advanced M3M-PCI64C myrinet cards, supports multiprogrammed semi-asynchronous checkpoints. The multiprogrammed approach allows higher degree of concurrency between checkpointing and other simulation specific operations carried out by the CPU, with obvious benefits on performance. We also report the results of the evaluation of those benefits for the case of a personal communication system simulation application.
Tasks running in computational Grids may require multiple types of resources simultaneously. Thus, co-allocation of multiple types of resources is needed when scheduling tasks. Traditional list scheduling heuristics c...
详细信息
Metacomputing is a method of using the GRID, which originated in the US and quickly was also picked up by European and Japanese researchers where a number of challenging projects were aiming at the exploitation of suc...
详细信息
Metacomputing is a method of using the GRID, which originated in the US and quickly was also picked up by European and Japanese researchers where a number of challenging projects were aiming at the exploitation of such distributed resources. Especially the GLOBUS project (The globus project: a status report, proceedings IPPS/SPDP'98 Heterogeneous Computing workshop, 1998, pp. 4-18) has contributed to the success of metacomputing substantially. However, high latency and low bandwidth have made people doubt the feasibility of this concept for big simulation codes. In this paper we present new numerical methods that help to exploit such configurations and overcome the problems of low network performance. To proof the feasibility of our approach we show results of simulations in an innovative GRID environment of supercomputers. (C) 2003 Elsevier Science (USA). All rights reserved.
In simulating large-scale networks, due to the limitation of available resources on computers, the size of the networks and the scale of simulation scenarios are often restricted Especially, routing tables, which indi...
详细信息
ISBN:
(纸本)0769519709
In simulating large-scale networks, due to the limitation of available resources on computers, the size of the networks and the scale of simulation scenarios are often restricted Especially, routing tables, which indicate the directions to forward packets, are considered to consume memory space. A simple general routing table requires O(N-2) space where N is the number of nodes. An algorithmic routing approach recently proposed by Heung et al. only requires O(N) space for representing routing tables, however this can be applied in the case that all the routes between two nodes are contained in a spanning tree (i.e. very limited routing strategies are allowed). In this paper we propose a new method to reduce the size of routing tables under any routing strategy. Given a general simple routing table, our method represents a routing table as the combination of an algorithmic routing based table and a general routing table, by translating a part of the given general routing table into the algorithmic routing based one. In order to reduce the size of the routing table, we find a (near-optimal)algorithmic routing based table that represents most part of the given routing table. Our experimental results have shown that our method could reduce the size of the table to 10% of the given routing table in hierarchical networks.
Recently, the scheme of,parallel downloading (PD) has been adopted by a number of Internet file downloading applications. With the wide deployment of content distribution networks and peer-to-peer networks, PD is expe...
详细信息
ISBN:
(纸本)0769519105
Recently, the scheme of,parallel downloading (PD) has been adopted by a number of Internet file downloading applications. With the wide deployment of content distribution networks and peer-to-peer networks, PD is expected to be more commonly used for file distribution. There have been experiments showing that PD results in higher aggregated downloading throughput and therefore shorter downloading time experienced by clients. However these experimental studies focused on the performance experienced by a particular user and did not consider the impact of PD on the network when it is largely deployed. In this paper we present our efforts toward an in-depth understanding of large-scale deployment of PD through simulation and analysis. Our results suggest that while PD may achieve a shorter downloading time, its impact on the network and server is significant. Our analysis is also used for network dimensioning and content distribution service provisioning. We show that with proper admission control and dimensioning, single-server downloading can perform just as well as PD, without the complexity and overhead incurred by PD.
The simulation of large-scale multicast networks often requires a significant amount of memory that can easily exceed the capacity of current computers, both because of the inherently large amount of state necessary t...
详细信息
We propose a new iterative distributed estimation algorithm for Gaussian hidden Markov graphical models with loops. We decompose a loopy graph into a number of linked embedded triangles and then apply a parallel block...
详细信息
ISBN:
(纸本)0780379977
We propose a new iterative distributed estimation algorithm for Gaussian hidden Markov graphical models with loops. We decompose a loopy graph into a number of linked embedded triangles and then apply a parallel block-Jacobi iteration comprising local linear minimum mean-square-error estimation on each triangle (involving a simple 3 x 3 matrix inverse computation) followed by an information exchange between neighboring nodes and triangles. A simulation study demonstrates that the algorithm converges extremely rapidly, outperforming a number of existing algorithms. Embedded triangles are simple, local, scalable, fault-tolerant, and energy-efficient, and thus ideally suited for wireless sensor networks.
Internet data traffic is doubling each year, yet bandwidth does not appear to be growing as fast as expected and thus short falls in available bandwidth, particularly at the "last mile" may result. To addres...
详细信息
Time-parallelsimulation defines a methodology that can be applied to certain specific simulation problems. We present a time-parallel approach for trace-driven simulation of the CSMA/CD protocol. The "memoryless...
详细信息
ISBN:
(纸本)9780769519708
Time-parallelsimulation defines a methodology that can be applied to certain specific simulation problems. We present a time-parallel approach for trace-driven simulation of the CSMA/CD protocol. The "memoryless" property of the physical system under moderate traffic loads allows for efficient time-parallelsimulation. We also present two optimization techniques: the estimation of idle points and the incremental fix-up computation. The former can improve the probability that a subtrace begins with a known initial system state. The latter can speedup the fix-up computation that is required when the estimation of the initial state is incorrect.
暂无评论