Keeping up with modern high-bandwidth networks is a significant challenge for system designers. A key obstacle to high network throughput is the high overhead of communication between the CPU and network interface controller (NIC), which typically resides on a standard I/O bus with high access latency. We investigate the impact of this overhead by analyzing the performance of hypothetical syste...

In most contemporary systems there are several jobs concurrently competing for shared resources, such as a processor, memory, network, sensors or other devices. Sharing a resource between several jobs requires synchronizing the jobs, specifying when which job will have access to the resource. A common synchronization method is scheduling. Executing a schedule requires switching resource assignm...

In concurrent computing environments based on heterogeneous processing elements interconnected by general-purpose networks, several classes of overheads contribute to lowered performance. The most obvious limitations are network throughput and latency, but certain other factors also play a significant role. In an attempt to gain some insight into the nature of these overheads, and to propose st...

This dissertation research addresses overheads in supporting virtual memory, especially virtual-to-physical address translation overheads (i.e., performance, power, and energy) via a Translation Lookaside Buffer (TLB). To overcome the overheads, we revisit virtually indexed, virtually tagged caches. In practice, they have not been common in commercial microarchitecture designs, and the crux of ...

To develop a good parallel implementation requires understanding of where run-time is spent and comparing this to some realistic best possible time. We introduce “overhead analysis” as a way of comparing achieved performance with achievable performance. We present a tool, OVALTINE, which aims to provide, automatically, the user with a hierarchical set of overheads for a given OpenMP implementat...

When threads use context switching, they incur an overhead in addition to the minimum required running time. The source of this overhead is both direct overhead due to running the context switch code and indirect overhead due to perturbation of caches. We calculate indirect overhead by measuring the running time of tasks that use context switching and subtracting the direct overhead. We also me...

| Overheads due to synchronisation and loop scheduling are an important factor in determining the performance of shared memory parallel programs. We present set of benchmarks to measure these classes of overhead for language constructs in OpenMP. Results are presented for three diierent hardware platforms, each with its own implementation of OpenMP. Signiicant diierences are observed, which sug...

Communication overheads are one of the fundamental challenges in a multiprocessor system. As the number of processors on a chip increases, communication overheads and the distribution of computation and data become increasingly significant. The granularity of communication between processors on a single chip in future systems will determine how significant these overheads will be. Fine-grained ...