Most multiprocessors employ coherent caches despite the overheads of doing so. As future processors will be multiprocessors with elaborate cache hierarchies, the overheads of cache coherence will be an important area for efficiency improvements. This paper proposes a novel technique, called Multi-Line Invalidation (MLI) messages, to reduce an important aspect of coherence overhead—the invalidat...

For fine-grain computation to be ejjective, the cost of communications between the large number of subtasks should be minimized. In this papec we present an optimization technique which reduces overheads of communications between local subtasks by bypassing the network inte$ace and transferring data directly from memory or registers to memory. On average, the optimization results in 35.6% impro...

Scientific workflows in High Performance Computing ( HPC ) environments are processing large amounts of data. The storage hierarchy on systems is getting deeper, driven by new technologies (NVRAMs, SSDs, etc.) There a need for programming abstractions that allow users to seamlessly manage data at the workflow level multi-tiered systems, and provide optimal performance use resources. In previous...

The use of parallelism enhances the performance of a software system. However, its excessive use can degrade the system performance. In this report we propose a parallelism viewpoint to optimize the use of parallelism by eliminating unnecessarily used parallelism in legacy systems. The parallelism viewpoint describes parallelism of the system in order to analyze multiple overheads associated wi...

This paper proposes an integrated network management framework for inter-domain outbound traffic engineering. The framework consists of three functional blocks (monitoring, optimization and implementation) to make the outbound traffic engineering adaptive to network condition changes such as inter-domain traffic demand variation, inter-domain routing changes and link failures. The objective is ...

Available network bandwidth is an important metrics of the network performance. Quickly and accurately measuring the available network bandwidth is significant to network monitoring, congestion control, and network traffic projects. The traditional Pathload scheme for measuring the available network bandwidth suffers heavy measurement overheads and slow convergence. Based on the analysis of Pat...

We offer a new metric for big data platforms, COST, or the Configuration that Outperforms a Single Thread. The COST of a given platform for a given problem is the hardware configuration required before the platform outperforms a competent single-threaded implementation. COST weighs a system’s scalability against the overheads introduced by the system, and indicates the actual performance gains ...

This paper presents an analytical model to evaluate the performance of conservative pamllel simulation using a finite-buffered multistage interconnection network as an example. The model is formalized based on two important time components in parallel and distributed processing: computation time and communication time. Our mathematical analysis identifies the major constituents of simulation ov...