Open Access. Powered by Scholars. Published by Universities.®
- Discipline
Articles 1 - 5 of 5
Full-Text Articles in OS and Networks
Statistical Modeling Of Hpc Performance Variability And Communication, Jered B. Dominguez-Trujillo
Statistical Modeling Of Hpc Performance Variability And Communication, Jered B. Dominguez-Trujillo
Computer Science ETDs
Understanding the performance of parallel and distributed programs remains a focal point in determining how compute systems can be optimized to achieve exascale performance. Lightweight, statistical models allow developers to both characterize and predict performance trade-offs, especially as HPC systems become more heterogeneous with many-core CPUs and GPUs. This thesis presents a lightweight, statistical modeling approach of performance variation which leverages extreme value theory by focusing on the maximum length of distributed workload intervals. This approach was implemented in MPI and evaluated on several HPC systems and workloads. I then present a performance model of partitioned communication which also uses …
Adaptive Parallelism For Coupled, Multithreaded Message-Passing Programs, Samuel K. Gutiérrez
Adaptive Parallelism For Coupled, Multithreaded Message-Passing Programs, Samuel K. Gutiérrez
Computer Science ETDs
Hybrid parallel programming models that combine message passing (MP) and shared- memory multithreading (MT) are becoming more popular, especially with applications requiring higher degrees of parallelism and scalability. Consequently, coupled parallel programs, those built via the integration of independently developed and optimized software libraries linked into a single application, increasingly comprise message-passing libraries with differing preferred degrees of threading, resulting in thread-level heterogeneity. Retroactively matching threading levels between independently developed and maintained libraries is difficult, and the challenge is exacerbated because contemporary middleware services provide only static scheduling policies over entire program executions, necessitating suboptimal, over-subscribed or under-subscribed, configurations. In …
Next Generation Tcp/Ip Side Channels, Xu Zhang
Next Generation Tcp/Ip Side Channels, Xu Zhang
Computer Science ETDs
Side channel techniques have been developed in recent years to fulfill various tasks in modern computer network measurements. However, due to their nature, these techniques are typically limited in terms of both fidelity and their ability to be used on the real Internet without raising ethical concerns because of packet rates. I propose the next generation of TCP/IP side channel techniques that exploit information flow in modern systems’ network stacks to overcome weaknesses in previous techniques. The proposed work is novel, non-intrusive, and can carry out measurements with high fidelity. I achieved this by deeply understanding the behaviors of modern …
Improving Hpc Communication Library Performance On Modern Architectures, Matthew G. F. Dosanjh
Improving Hpc Communication Library Performance On Modern Architectures, Matthew G. F. Dosanjh
Computer Science ETDs
As high-performance computing (HPC) systems advance towards exascale (10^18 operations per second), they must leverage increasing levels of parallelism to achieve their performance goals. In addition to increased parallelism, machines of that scale will have strict power limitations placed on them. One direction currently being explored to alleviate those issues are many-core processors such as Intel’s Xeon Phi line. Many-core processors sacrifice clock speed and core complexity, such as out of order pipelining, to increase the number of cores on a die. While this increases floating point throughput, it can reduce the performance of serialized, synchronized, and latency sensitive code …
Characterizing And Improving Power And Performance In Hpc Networks, Taylor L. Groves
Characterizing And Improving Power And Performance In Hpc Networks, Taylor L. Groves
Computer Science ETDs
Networks are the backbone of modern HPC systems. They serve as a critical piece of infrastructure, tying together applications, analytics, storage and visualization. Despite this importance, we have not fully explored how evolving communication paradigms and network design will impact scientific workloads. As networks expand in the race towards Exascale (1×10^18 floating point operations a second), we need to reexamine this relationship so that the HPC community better understands (1) characteristics and trends in HPC communication; (2) how to best design HPC networks to save power or enhance the performance; (3) how to facilitate scalable, informed, and dynamic decisions within …