Nikitas J. Dimopoulos

A new heuristic for solving the multichoice multidimensional knapsack problem

A new heuristic for solving the multichoice multidimensional knapsack problem (MMKP) is presented in this paper. The MMKP is first reduced to a multidimensional knapsack problem (MKP). A linear programming relaxation of the resulting MKP is solved, and a series of new values for the variables is computed. These values, pseudo-utility values, and resource value coefficients computed as well, are used in order to obtain a feasible solution for the original MMKP. Finally, the quality of the feasible solution is improved using the pseudo-utility values and the coefficient values of the objective function. Numerical results show that the performance of this approach is superior to that of previous techniques.

Resource allocation on computational grids using a utility model and the knapsack problem

This work introduces a utility model (UM) for resource allocation on computational grids and formulates the allocation problem as a variant of the 0-1 multichoice multidimensional knapsack problem. The notion of task-option utility is introduced, and it is used to effect allocation policies. We present a variety of allocation policies, which are expressed as functions of metrics that are both intrinsic and external to the task and resources. An external user-defined credit-value metric is shown to allow users to intervene in the allocation of urgent or low priority tasks. The strategies are evaluated in simulation against random workloads as well as those drawn from real systems. We measure the sensitivity of the UM-derived schedules to variations in the allocation policies and their corresponding utility functions. The UM allocation strategy is shown to optimally allocate resources congruent with the chosen policies.

Efficient communication using message prediction for clusters of multiprocessors

With the increasing uniprocessor and symmetric multiprocessor computational power available today, interprocessor communication has become an important factor that limits the performance of clusters of workstations/multiprocessors. Many factors including communication hardware overhead, communication software overhead, and the user environment overhead (multithreading, multiuser) affect the performance of the communication subsystems in such systems. A significant portion of the software communication overhead belongs to a number of message copying operations. Ideally, it is desirable to have a true zero‐copy protocol where the message is moved directly from the send buffer in its user space to the receive buffer in the destination without any intermediate buffering. However, due to the fact that message‐passing applications at the send side do not know the final receive buffer addresses, early arrival messages have to be buffered at a temporary area. In this paper, we show that there is a message reception communication locality in message‐passing applications. We have utilized this communication locality and devised different message predictors at the receiver sides of communications. In essence, these message predictors can be efficiently used to drain the network and cache the incoming messages even if the corresponding receive calls have not yet been posted. The performance of these predictors, in terms of hit ratio, on some parallel applications are quite promising and suggest that prediction has the potential to eliminate most of the remaining message copies. We also show that the proposed predictors do not have sensitivity to the starting message reception call, and that they perform better than (or at least equal to) our previously proposed predictors. Copyright

A study of the asymptotic behavior of neural networks

The stability properties are studied of neural networks modeled as a set of nonlinear differential equations of the form TX+X=Wf(X)+b where X is the neural membrane potential vector, W is the network connectivity matrix, and F(X) is the nonlinearity (an essentially sigmoid function). Topologies of neural networks that exhibit asymptotic behavior are established. This behavior depends solely on the topology of the network. Moreover, the connectivity W need not be symmetric. Networks topologically similar to the cerebellum fall in this category and exhibit asymptotic behavior. The simulated behavior of typical neural networks is presented. >

A theory for total exchange in multidimensional interconnection networks

Total exchange (or multiscattering) is one of the important collective communication problems in multiprocessor interconnection networks. It involves the dissemination of distinct messages from every node to every other node. We present a novel theory for solving the problem in any multidimensional (cartesian product) network. These networks have been adopted as cost-effective interconnection structures for distributed-memory multiprocessors. We construct a general algorithm for single-port networks and provide conditions under which it behaves optimally. It is seen that many of the popular topologies, including hypercubes, k-ary n-cubes, and general tori satisfy these conditions. The algorithm is also extended to homogeneous networks with 2/sup k/ dimensions and with multiport capabilities. Optimality conditions are also given for this model. In the course of our analysis, we also derive a formula for the average distance of nodes in multidimensional networks; it can be used to obtain almost closed-form results for many interesting networks.

Resource management and knapsack formulations on the grid

This work formulates the resource allocation problem on grids as a knapsack problem. The notion of utility is introduced, and it is used to effect allocation policies. Simulation results using a variety of allocation policies are presented and show that knapsack formulations optimally allocate resources congruent with the chosen policies.

Efficient Communication Using Message Prediction for Cluster Multiprocessors

With the increasing uniprocessor and SMP computation power available, interprocessor communication has become an important factor that limits the performance of clusters of workstations. Many factors including communication hardware overhead, communication software overhead, and the user environment overhead (multithreading, multiuser) affect the performance of the communication subsystems. A significant portion of the software communication overhead is attributed to message copying. Ideally, it is desirable to have a true zero-copy protocol where the message is moved directly from the send buffer in its user space to the receive buffer in the destination. However, because the send side does not know the final receive buffer address, early arriving messages have to be buffered at a temporary area. In this work, we show that there is a message reception communication locality in message-passing applications. We have utilized this communication locality and devised different message predictors at the receiver sides of communications. In essence, these message predictors can be used to drain the network and cache the incoming messages even if the corresponding receive calls have not been posted yet. The performance of these predictors, in terms of hit ratio, on some parallel applications is quite promising and suggest that prediction has the potential to eliminate most of the remaining message copies.

Metascheduling Multiple Resource Types using the MMKP

Grid computing involves the transparent sharing of computational resources of many types by users across large geographic distances. The altruistic nature of many current grid resource contributions does not encourage efficient usage of resources. As grid projects mature, increased resource demands coupled with increased economic interests will introduce a requirement for a metascheduler that improves resource utilization, allows administrators to define allocation policies, and provides an overall quality of service to the grid users. In this work we present one such metascheduling framework, based on the multichoice multidimensional knapsack problem (MMKP). This strategy maximizes overall grid utility by selecting desirable options of each task subject to constraints of multiple resource types. We present the framework for the MMKP metascheduler and discuss a selection of allocation policies and their associated utility functions. The MMKP metascheduler and allocation policies are demonstrated using a grid of processor, storage, and network resources. In particular, a data transfer time metric is incorporated into the utility function in order to prefer task options with the lowest data transfer times. The resulting schedules are shown to be consistent with the defined policies

Extended characterization of DMA transfers on the Cell BE processor

The main contributors to message delivery latency in message passing environments are the copying operations needed to transfer and bind a received message to the consuming process/thread. A significant portion of the software communication overhead is attributed to message copying. Recently, a set of factors has been leading high- performance processor architectures toward designs that feature multiple processing cores on a single chip (a.k.a. CMP). The Cell Broadband Engine (BE) shows potential to provide high-performance to parallel applications (e.g., MPI applications). The Cells non-homogeneous architecture along with small local storage in SPEs impose restrictions and challenges for parallel applications. In this work, we first characterize various data delivery mechanisms in the Cell BE processor; then, we propose techniques to facilitate the delivery of a message in MPI environments implemented in the Cell BE processor. We envision a cluster system comprising several cell processors each supporting several computation threads.

Lazy direct-to-cache transfer during receive operations in a message passing environment

The focus of this work is on techniques that promise to reduce the message delivery latency in message passing interface (MPI) environments. The main contributors to message delivery latency in message passing environments are the copying operations needed to transfer and bind a received message to the consuming process/thread. To reduce this copying overhead and to reach toward finer granularity, we introduce architectural extensions comprising of a specialized network cache and instructions to manage the operations of this extension. In this work we study the caching environment and evaluate a new technique called Lazy Direct-to-Cache Transfer (DTCT). Our simulations show that messages can be bound and kept into a network cache where they persist long enough to be consumed. We also demonstrate that lazy DTCT provides a significant reduction in the access latency for I/O intensive environments such as message passing configurations and SMPs without polluting the data cache.