The mining of multi-dimensional time series is a crucial step in gaining insights into data obtained from physical systems and from monitoring infrastructures. A widely accepted approach for this challenge is the matrix profile, which, however, is computationally very expensive. It relies on calculating large correlation matrices coupled with sort operations across all dimensions of the data, as well as on performing inclusive scans. All of these steps are inherently data parallel and can, therefore, benefit from execution on GPUs, and even more so from horizontal scaling on multiple GPUs. In addition, the nature of the matrix profile calculation allows the exploitation of reduced precision on GPUs. This offers further improvements to enable the analysis of ever growing data sets in real-world scenarios. Based on these motivations, we introduce the first parallel algorithm for multi-dimensional matrix profile on multiple GPUs exploiting reduced precision modes and provide a highly opti-mized implementation using novel optimization techniques. On one NVIDIA A100 GPU, our implementation achieves a 54x performance improvement in comparison to an optimized single-node execution on a state-of-the-art CPU-based implementation relying on double-precision computation and an additional factor of 1.4x when switching to reduced precision while maintaining sufficient accuracy. We study the accuracy and performance trade-offs for our proposed algorithm in detail and present synthetic and real-world case studies to demonstrate how the reduced precision improves the performance, while accomplishing sufficiently accurate results.

Future HPC applications must be able to scale to thousands of compute nodes, while running for several days. The increased runtime and node count inconveniently raises the probability of hardware failures that may interrupt computations. Scientists must therefore protect their simulations against hardware failures. This is typically done using frequent checkpoint& restart, which may have significant overheads. Consequently, the frequency in which checkpoints are taken should be minimized. Predicting hardware failures ahead of time is a promising approach to address this problem, but has remaining issues like false alarms at large scales. In this paper, we introduce the probability of unnecessarily triggering checkpoints (UC) to evaluate the quality of node level failure predictors for checkpointing large-scale applications. This metric is used to show how current predictors suffer from too many false alarms at large node counts. Further, we propose a new failure predictor that chains several machine learning classifiers to make predictions with minimal false alarms. We aim for extremely low false positive rates to guarantee that no unnecessary checkpoints will be performed even for very large node counts. Our experiments based on real system traces from a large production cluster show that our predictor achieves a lead-up time of four minutes, a recall of 0.7302, a false positive rate of 0.0004, a precision of 0.9944 and a probability of unnecessary checkpoints (UC) of 0.00011 for 1024 nodes.

As we approach exascale and start planning for beyond, the rising complexity of systems and applications demands new monitoring, analysis, and optimization approaches. This requires close coordination with the parallel programming system used, which for HPC in most cases includes MPI, the Message Passing Interface. While MPI provides comprehensive tool support in the form of the MPI Profiling interface, PMPI, which has inspired a generation of tools, it is not sufficient for the new arising challenges. In particular, it does not support modern software design principles nor the composition of multiple monitoring solutions from multiple agents or sources. We approach these gaps and present QMPI, as a possible successor to PMPI. In this paper, we present the use cases and requirements that drive its development, offer a prototype design and implementation, and demonstrate its effectiveness and low overhead.

As we near exascale, resilience remains a major technical hurdle. Any technique with the goal of achieving resilience suffers from having to be reactive, as failures can appear at any time. A wide body of research aims at predicting failures, i.e., forecasting failures so that evasive actions can be taken while the system is still fully functional, which has the benefit of giving insight into the global system state.

Two extension tools for enhancing the compression performance of prediction-based lossless audio coding are proposed. One is progressive-order prediction of the starting samples at the random access points, where the information of previous samples is not available. The first sample is coded as is, the second is predicted by first-order prediction, the third is predicted by second-order prediction, and so on. This can be efficiently carried out with PAR-COR (PARtial autoCORrelation) coefficients. The second tool is interchannel joint coding. Both predictive coefficients and prediction error signals are efficiently coded by interchannel differential or three-tap adaptive prediction. These new prediction tools lead to a steady reduction in bit rate when random access is activated and the interchannel correlation is strong.

We propose an efficient lossless coding algorithm that not only handles both PCM format data and IEEE floating-point format data, but also provides end users with a random access property. In the worst-case scenario, where the proposed algorithm was applied to artificially generated full 32 bit floating-point sound files with 48 kHz or 96 kHz sampling frequencies, an average compression rate of more than 1.5 and 1.7, respectively, was still achieved, which is much better than the average compression rate of less than 1.1 achieved by the general purpose lossless coding algorithm, gzip. Moreover, input sound files with samples' magnitudes out-of-range can also be perfectly reconstructed by our algorithm.

A new quality-scalable high-fidelity multichannel audio compression algorithm based on MPEG-2 advanced audio coding (AAC) is presented. The Karhunen-Loeve transform (KLT) is applied to multichannel audio signals in the preprocessing stage to remove interchannel redundancy. Then, signals in decorrelated channels are compressed by a modified AAC main profile encoder. Finally, a channel transmission control mechanism is used to re-organize the bitstream so that the multichannel audio bitstream has a quality scalable property when it is transmitted over a heterogeneous network. Experimental results show that, compared with AAC, the proposed algorithm achieves a better performance while maintaining a similar computational complexity at the regular bit rate of 64 kbit/sec/ch. When the bitstream is transmitted to narrowband end users at a lower bit rate, packets in some channels can be dropped, and slightly degraded, yet full-channel, audio can still be reconstructed in a reasonable fashion without any additional computational cost.

MPEG Advanced Audio Coding (AAC) is one of the most distinguished multichannel digital audio compression systems. Based on AAC, we develop a progressive syntax-rich multichannel audio codec in this work. It not only supports fine grain bit rate scalability for multichannel audio bitstreams but also provides several desirable functionalities. MPEG-4 version-2 audio coding supports fine grain bit rate scalablility in its Generic Audio Coder with the Bit-Sliced Arithmetic Coding (BSAC) tool that provides scalability for mono or stereo audio, but not for multichannel audio. A formal subjective listening test shows that, the proposed algorithm achieves a better performance at several different bit rates when compared with MPEG-4 BSAC for mono audio sources.