Conference
High Performance Computing Symposium
About: High Performance Computing Symposium is an academic conference. The conference publishes majorly in the area(s): Computer science & Parallel algorithm. Over the lifetime, 144 publications have been published by the conference receiving 856 citations.
Papers published on a yearly basis
Papers
9 Feb 2012
TL;DR: Using a recently developed chemistry model for ablating carbon-phenolic-in-air species, a CFD calculation of the Stardust re-entry at 71 km is presented and the predicted emission from the CN lines compares quite well with the experimental results, demonstrating the validity of the current approach.
Abstract: Re-entry vehicles designed for space exploration are usually equipped with thermal protection systems made of ablative material. In order to properly model and predict the aerothermal environment of the vehicle, it is imperative to account for the gases produced by ablation processes. In the case of charring ablators, where an inner resin is pyrolyzed at a relatively low temperature, the composition of the gas expelled into the boundary layer is complex and may lead to thermal chemical reactions that cannot be captured with simple flow chemistry models. In order to obtain better predictions, an appropriate gas flow chemistry model needs to be included in the CFD calculations. Using a recently developed chemistry model for ablating carbon-phenolic-in-air species, a CFD calculation of the Stardust re-entry at 71 km is presented. The code used for that purpose has been designed to take advantage of the nature of the problem and therefore remains very efficient when a high number of chemical species are involved. The CFD result demonstrates the need for such chemistry model when modeling the flow field around an ablative material. Modeling of the nonequilibrium radiation spectra is also presented, and compared to the experimental data obtained during Stardust re-entry by the Echelle instrument. The predicted emission from the CN lines compares quite well with the experimental results, demonstrating the validity of the current approach.
101 citations
7 Apr 2013
TL;DR: Experiments show that the overall system efficiency and availability would go towards zero as system scales approach exascale with checkpointing mechanism on parallel filesystems, however, the simulations suggest that a distributed filesystem with local persistent storage would offer excellent scalability and aggregate bandwidth, enabling efficient checkpointing at exASCale.
Abstract: Exascale computers are predicted to emerge by the end of this decade with millions of nodes and billions of concurrent cores/threads. One of the most critical challenges for exascale computing is how to effectively and efficiently maintain the system reliability. Checkpointing is the state-of-the-art technique for high-end computing system reliability that has proved to work well for current petascale scales. This paper investigates the suitability of checkpointing mechanism for exascale computers, across both parallel filesystems and distributed filesystems. We built a model to emulate exascale systems, and developed a simulator, RXSim, to study its reliability and efficiency. Experiments show that the overall system efficiency and availability would go towards zero as system scales approach exascale with checkpointing mechanism on parallel filesystems. However, the simulations suggest that a distributed filesystem with local persistent storage would offer excellent scalability and aggregate bandwidth, enabling efficient checkpointing at exascale.
29 citations
13 Apr 2014
TL;DR: This paper implements a solution that uses "remote API execution" and takes advantage of DirectPath I/O to enable general purpose GPU on ESX to enable CUDA applications running concurrently in multiple VMs onESX to share GPU(s).
Abstract: Graphics Processing Units (GPU) have become important components in high performance computing (HPC) systems for their massively parallel computing capability and energy efficiency. Virtualization technologies are increasingly applied to HPC to reduce administration costs and improve system utilization. However, virtualizing the GPU to support general purpose computing presents many challenges because of the complexity of this device. On VMware's ESX hypervisor, DirectPath I/O can provide virtual machines (VM) high performance access to physical GPUs. However, this technology does not allow multiplexing for sharing GPUs among VMs and is not compatible with vMotion, VMware's technology for transparently migrating VMs among hosts inside clusters. In this paper, we address these issues by implementing a solution that uses "remote API execution" and takes advantage of DirectPath I/O to enable general purpose GPU on ESX. This solution, named vmCUDA, allows CUDA applications running concurrently in multiple VMs on ESX to share GPU(s). Our solution requires neither recompilation nor even editing of the source code of CUDA applications. Our performance evaluation has shown that vmCUDA introduced an overhead of 0.6% - 3.5% for applications with moderate data size and 14% - 20% for those with large data (e.g. 12.5 GB - 237.5GB in our experiments).
27 citations
7 Apr 2013
TL;DR: A light-weight discrete event simulator, SimMatrix, which simulates job scheduling system comprising of millions of nodes and billions of cores/tasks, and which validated against two real systems and compared with SimGrid and GridSim in terms of resource consumption at scale.
Abstract: Exascale computers (expected to be composed of millions of nodes and billions of threads of execution) will enable the unraveling of significant scientific mysteries. Many-task computing is a distributed paradigm, which can potentially address three of the four major challenges of exascale computing, namely Memory/Storage, Concurrency/Locality, and Resiliency. Exascale computing will require efficient job scheduling/management systems that are several orders of magnitude beyond the state-of-the-art, which tend to have centralized architecture and are relatively heavy-weight. This paper proposes a light-weight discrete event simulator, SimMatrix, which simulates job scheduling system comprising of millions of nodes and billions of cores/tasks. SimMatrix supports both centralized (e.g. first-in-first-out) and distributed (e.g. work stealing) scheduling. We validated SimMatrix against two real systems, Falkon and MATRIX, with up to 4K-cores, running on an IBM Blue Gene/P system, and compared SimMatrix with SimGrid and GridSim in terms of resource consumption at scale. Results show that SimMatrix consumes up to two-orders of magnitude lower memory per task, and at least one-order of magnitude (and up to four-orders of magnitude) lower time per task overheads. For example, running a workload of 10 billion tasks on 1 million nodes and 1 billion cores required 142GB memory and 163 CPU-hours. These relatively low costs at exascale levels of concurrency will lead to innovative studies in scheduling algorithms at unprecedented scales.
25 citations
3 Apr 2016
TL;DR: DrAFTS is presented -- a methodology for predicting the value a user should bid in the AWS Spot tier to ensure that the request lifetime exceeds a fixed duration with a given probability.
Abstract: In this paper, we present DrAFTS -- a methodology for predicting the value a user should bid in the AWS Spot tier to ensure that the request lifetime exceeds a fixed duration with a given probability. DrAFTS uses previous price histories for each category of request to determine both the minimum bid value and the concomitant duration associated with that value that is "guaranteed" with a specified probability. We also describe how DrAFTS can be used to determine the duration associated with over-bidding the minimum by a specific percentage.
23 citations
Performance Metrics
| Year | Papers |
|---|---|
| 2019 | 2 |
| 2018 | 13 |
| 2017 | 17 |
| 2016 | 47 |
| 2015 | 1 |
| 2014 | 25 |