Conventional RAID solutions with fixed layouts partition large disk enclosures so that each RAID group uses its own disks exclusively. This achieves good performance isolation across underlying disk groups, at the cost of disk under-utilization and slow RAID reconstruction from disk failures. We propose RAID+, a new RAID construction mechanism that spreads both normal I/O and reconstruction workloads to a larger disk pool in a balanced manner. Unlike systems conducting randomized placement, RAID+ employs deterministic addressing enabled by the mathematical properties of mutually orthogonal Latin squares, based on which it constructs 3-D data templates mapping a logical data volume to uniformly distributed disk blocks across all disks. While the total read/write volume remains unchanged, with or without disk failures, many more disk drives participate in data service and disk reconstruction. Our evaluation with a 60-drive disk enclosure using both synthetic and real-world workloads shows that RAID+ significantly speeds up data recovery while delivering better normal I/O performance and higher multi-tenant system throughput.

Existing GPU-based approaches cannot yet meet the performance requirement for training very large convolutional neural networks (CNNs), where convolutional layers (Conv-layers) dominate the training time. In this paper, we find that no single convolution policy can always perform the fastest across all the computing phases. Then, we propose an approach called HyConv to accelerating multi-phase CNN computation by fine-grained policy selection. HyConv encapsulates existing convolution policies into a set of modules, and selects the fastest policy (a.k.a., winner policy) via one-round runtime measurement for computing each phase. Furthermore, HyConv uses a winner database to record the current winner policies, avoiding duplicate measurement later for the same parameter configuration. Our experimental results indicate that over all the used real-world CNN networks, HyConv consistently outperforms existing approaches on either a single GPU or four GPUs, with speedups of up to 3.3× and up to 1.6× over cuDNN-MM respectively. Such improvement can be explained by our result that HyConv delivers obviously better performance for most of single Conv-layers. Furthermore, HyConv has the ability to work with any parameter configuration and thus keeps better usability.

As one of the most important deep learning models, convolutional neural networks (CNNs) have achieved great successes in a number of applications such as image classification, speech recognition and nature language understanding. Training CNNs on large data sets is computationally expensive, leading to a flurry of research and development of open-source parallel implementations on GPUs. However, few studies have been performed to evaluate the performance characteristics of those implementations. In this paper, we conduct a comprehensive comparison of these implementations over a wide range of parameter configurations, investigate potential performance bottlenecks and point out a number of opportunities for further optimization.