Accurate and timely regional weather prediction is vital for sectors dependent on weather-related decisions. Traditional prediction methods, based on atmospheric equations, often struggle with coarse temporal resolutions and inaccuracies. This article presents a novel machine learning (ML) model, called Micro-Macro (MiMa), that integrates both near-surface observational data from Kentucky Mesonet stations (collected every 5 min, known as Micro data) and hourly atmospheric numerical outputs (termed as Macro data) for fine-resolution weather forecasting. The MiMa model employs an encoder-decoder transformer structure, with two encoders for processing multivariate data from both datasets and a decoder for forecasting weather variables over short time horizons. Each instance of the MiMa model, called a modelet, predicts the values of a specific weather parameter at an individual mesonet station. The approach is extended with Regional MiMa (Re-MiMa) modelets, which are designed to predict weather variables at ungauged locations by training on multivariate data from a few representative stations in a region, tagged with their elevations. Re-MiMa can provide highly accurate predictions across an entire region, even in areas without observational stations. Experimental results show that MiMa significantly outperforms current models, with Re-MiMa offering precise short-term forecasts for ungauged locations, marking a significant advancement in weather forecasting accuracy and applicability.

The proliferation of IoT devices, with various capabilities in sensing, monitoring, and controlling, has prompted diverse emerging applications, highly relying on effective delivery of sensitive information gathered at edge devices to remote controllers for timely responses. To effectively deliver such information/status updates, this paper undertakes a holistic study of AoI in multi-hop networks by considering the relevant and realistic factors, aiming for optimizing information freshness by rapidly shipping sensitive updates captured at a source to its destination. In particular, we consider the multi-channel with OFDM (orthogonal frequency-division multiplexing) spectrum access in multi-hop networks and develop a rigorous mathematical model to optimize AoI at destination nodes. Real-world factors, including orthogonal channel access, wireless interference, and queuing model, are taken into account for the very first time to explore their impacts on the AoI. To this end, we propose two effective algorithms where the first one approximates the optimal solution as closely as we desire while the second one has polynomial time complexity, with a guaranteed performance gap to the optimal solution. The developed model and algorithms enable in-depth studies on AoI optimization problems in OFDM-based multi-hop wireless networks. Numerical results demonstrate that our solutions enjoy better AoI performance and that AoI is affected markedly by those realistic factors taken into our consideration.

Recent graphic processing units (GPUs) often come with large on-board physical memory to accelerate diverse parallel program executions on big datasets with regular access patterns, including machine learning (ML) and data mining (DM). Such a GPU may underutilize its physical memory during lengthy ML model training or DM, making it possible to lend otherwise unused GPU memory to applications executed concurrently on the host machine. This work explores an effective approach that lets memory-intensive applications run on the host machine CPU with its memory expanded dynamically onto available GPU on-board DRAM, called GPU-assisted memory expansion (GAME). Targeting computer systems equipped with the recent GPUs, our GAME approach permits speedy executions on CPU with large memory footprints by harvesting unused GPU on-board memory on-demand for swapping, far surpassing competitive GPU executions. Implemented in user space, our GAME prototype lets GPU memory house swapped-out memory pages transparently, without code modifications for high usability and portability. The evaluation of NAS-NPB benchmark applications demonstrates that GAME expedites monotasking (or multitasking) executions considerably by up to 2.1× (or 3.1×), when memory footprints exceed the CPU DRAM size and an equipped GPU has unused VDRAM available for swapping use.

This article considers best checkpointing control realizable in real-world systems, whose mean time between failures (MTBFs) often fluctuate. The considered control scheme is based on equating aggregate checkpointing overhead over an activity sequence of interest (θ) and the expected rework amount after a failure recovery for best checkpointing, called “CHORE” (i.e., checkpointing overhead and rework equated), where θ starts from execution resumption after failure recovery and ends after restore from the following failure. CHORE lets its inter-checkpoint intervals in θ follow a pre-determined sequence independent of MTBF to aim at performance optimality and is shown analytically to keep overall execution time overhead upper bounded. When failure occurrences are tracked during job execution for real-time MTBF estimation, an enhanced CHORE (dubbed En-CHORE) is obtained to lower checkpointing overhead by skipping certain checkpoints at the beginning of each θ before taking checkpoints with the most desirable inter-checkpoint intervals determined on-the-fly for best checkpointing control. En-CHORE can outperform optimal checkpointing (which follows a fixed inter-checkpoint interval optimized for one constant global MTBF known a prior) both under synthetic random failures with local MTBF fluctuating markedly and under real failure traces of 22 real HPC systems (whose failure rates actually fluctuate over their trace time spans).

A bufferless network-on-chip (NoC) can deliver high energy efficiency, but such a NoC is subject to growing deflection when its traffic load rises. This article proposes Deflection Containment (DeC) for the bufferless NoC to address its notorious shortcomings of excessive deflection for performance improvement and energy savings. With multiple subnetworks bridged by an added link between two corresponding routers, DeC lets a contending flit in one subnetwork be forwarded to another subnetwork instead of deflected. Microarchitecture of DeC routers is rectified to shorten the critical path and lift network bandwidth. Its Cadence RTL implementations with a 15 - nm process are conducted respectively for mesh-based NoCs and torus-based NoCs. Additionally, different sized DeC-NoCs are evaluated extensively and compared with previous bufferless designs (BLESS and MinBD), uncovering that DeC with two bridged subnetworks (dubbed DeC2) for 8×8 mesh-based NoCs can lower deflection drastically by some 90 percent and energy consumption by upto 51 percent under real benchmark traffic loads, in comparison to BLESS. Under various synthetic traffic models and workloads, 16×16 torus-based DeC2-NoC sustains up to 2.33× loads when compared with its mesh-based counterpart, exhibiting the same clock rate and taking only negligible more power and area according to our full layout results.

In multicore systems, a large portion of checkpoint time overhead can be hidden from the execution critical path by resorting to a dedicated checkpointing thread run concurrently with regular execution threads for compressing checkpoint files to lower checkpointing overhead. On the other hand, the restore time is on the critical path that cannot be hidden, making it most important to accelerate execution restore upon failures. This work pursues a restore-express (REX) strategy for multi-level checkpointing (MLC), applicable to any incremental checkpointing (IC). Oblivious to application codes, REX employs adaptive IC (AIC) for local (L1) checkpointing and follows our runtime control for second-level (L2) checkpointing, with its aim at express restore from failures while holding down the overall execution time. It takes advantage of two unique insights for overhead reduction: (1) the modified pages of an incremental checkpoint file are likely to exist in a subsequent checkpoint file, and (2) many data patterns (on an average, some 40 percent of them) stay unchanged from one L2 checkpoint file to the next. These insights enable REX to (1) coalesce IC files (by involving only the last copy of every dirty page among files) and (2) boost file compression across multiple L2 checkpoints. Time and storage overhead results of REX during normal job execution are gathered for 16 benchmarks from SPEC, PARSEC, and NPB suites. The evaluation outcomes of the execution restore time confirm that REX is fast and able to quicken restore by a factor of 4.5× when compared with its IC counterpart (without utilizing the unique insights), while incurring same execution time overhead.