Engineering topics

Kang, Qiao

Publications and source records attributed to Kang, Qiao.

3 records

DOE OSTI2022

Data Storage for HEP Experiments in the Era of High-Performance Computing

As particle physics experiments push their limits on both the energy and the intensity frontiers, the amount and complexity of the produced data are also expected to increase accordingly. With such large data volumes, next-generation efforts like the HL-LHC and DUNE will rely even more on both high-throughput (HTC) and high-performance (HPC) computing clusters. Full utilization of HPC resources requires scalable and efficient data-handling and I/O. For the last few decades, ROOT has been used by most HEP experiments to store data. However, other storage technologies like HDF5 may perform better in HPC environments. Initial explorations with HDF5 have begun using ATLAS, CMS and DUNE data; the DUNE experiment has also adopted HDF5 for its data-acquisition system. This paper presents the future outlook of the HEP computing and the role of HPC, and a summary of ongoing and future works to use HDF5 as a possible data storage technology for the HEP experiments to use in HPC environments.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

DOE OSTI2021

Improving scalability of parallel CNN training by adaptively adjusting parameter update frequency

Synchronous SGD with data parallelism, the most popular parallelization strategy for CNN training, suffers from the expensive communication cost of averaging gradients among all workers. The iterative parameter updates of SGD cause frequent communications and it becomes the performance bottleneck. In this paper, we propose a lazy parameter update algorithm that adaptively adjusts the parameter update frequency to address the expensive communication cost issue. Our algorithm accumulates the gradients if the difference of the accumulated gradients and the latest gradients is sufficiently small. Here, the less frequent parameter updates reduce the per-iteration communication cost while maintaining the model accuracy. Our experimental results demonstrate that the lazy update method remarkably improves the scalability while maintaining the model accuracy. For ResNet50 training on ImageNet, the proposed algorithm achieves a significantly higher speedup (739.6 on 2048 Cori KNL nodes) as compared to the vanilla synchronous SGD (276.6) while the model accuracy is almost not affected (<0.2% difference).

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2020

Improving MPI Collective I/O for High Volume Non-Contiguous Requests With Intra-Node Aggregation

Two-phase I/O is a well-known strategy for implementing collective MPI-IO functions. It redistributes I/O requests among the calling processes into a form that minimizes the file access costs. As modern parallel computers continue to grow into the exascale era, the communication cost of such request redistribution can quickly overwhelm collective I/O performance. In this study, this effect has been observed from parallel jobs that run on multiple compute nodes with a high count of MPI processes on each node. Furthermore, to reduce the communication cost, we present a new design for collective I/O by adding an extra communication layer that performs request aggregation among processes within the same compute nodes. This approach can significantly reduce inter-node communication contention when redistributing the I/O requests. We evaluate the performance and compare it with the original two-phase I/O on Cray XC40 parallel computers (Theta and Cori) with Intel KNL and Haswell processors. Using I/O patterns from two large-scale production applications and an I/O benchmark, we show our proposed method effectively reduces the communication cost and hence maintains the scalability for a large number of processes.

97 MATHEMATICS AND COMPUTING↗