Performance Modeling in CUDA Streams - A Means for High-Throughput Data Processing

Dual-wavelength off-axis digital holography in ImageJ: toward real-time phase retrieval using CUDA streams

An ImageJ plug-in is developed to realize automatic real-time phase reconstruction for dual-wavelength digital holography (DH). This plug-in assembles the algorithms, including automatic phase reconstruction based on the division algorithm and post-processing. These algorithms are implemented and analyzed using a CPU and GPU, respectively. To hide the CPU-to-GPU data transfer latency, an optimization scheme using Compute Unified Device Architecture (CUDA) streams is proposed in ImageJ. Experimental results show that the proposed plug-in can perform faster reconstruction for dual-wavelength DH, resulting in frame rates up to 48 fps even for one-megapixel digital holograms on a normal PC. In other words, the proposed plug-in can realize real-time phase reconstruction for dual-wavelength digital holographic videos.

Dynamic Memory Management in Massively Parallel Systems: A Case on GPUs

Due to the high level of parallelism, there are unique challenges in developing system software on massively parallel hardware such as GPUs. One such challenge is designing a dynamic memory allocator whose task is to allocate memory chunks to requesting threads at runtime. State-of-the-art GPU memory allocators maintain a global data structure holding metadata to facilitate allocation/deallocation. However, the centralized data structure can easily become a bottleneck in a massively parallel system. In this paper, we present a novel approach for designing dynamic memory allocation without a centralized data structure. The core idea is to let threads follow a random search procedure to locate free pages. Then we further extend to more advanced designs and algorithms that can achieve an order of magnitude improvement over the basic idea. We present mathematical proofs to demonstrate that (1) the basic random search design achieves asymptotically lower latency than the traditional queue-based design and (2) the advanced designs achieve significant improvement over the basic idea. Extensive experiments show consistency to our mathematical models and demonstrate that our solutions can achieve up to two orders of magnitude improvement in latency over the best-known existing solutions.

Concurrent query processing in a GPU-based database system

The unrivaled computing capabilities of modern GPUs meet the demand of processing massive amounts of data seen in many application domains. While traditional HPC systems support applications as standalone entities that occupy entire GPUs, there are GPU-based DBMSs where multiple tasks are meant to be run at the same time in the same device. To that end, system-level resource management mechanisms are needed to fully unleash the computing power of GPUs in large data processing, and there were some researches focusing on it. In our previous work, we explored the single compute-bound kernel modeling on GPUs under NVidia's CUDA framework and provided an in-depth anatomy of the NVidia's concurrent kernel execution mechanism (CUDA stream). This paper focuses on resource allocation of multiple GPU applications towards optimization of system throughput in the context of systems. Comparing to earlier studies of enabling concurrent tasks support on GPU such as MultiQx-GPU, we use a different approach that is to control the launching parameters of multiple GPU kernels as provided by compile-time performance modeling as a kernel-level optimization and also a more general pre-processing model with batch-level control to enhance performance. Specifically, we construct a variation of multi-dimensional knapsack model to maximize concurrency in a multi-kernel environment. We present an in-depth analysis of our model and develop an algorithm based on dynamic programming technique to solve the model. We prove the algorithm can find optimal solutions (in terms of thread concurrency) to the problem and bears pseudopolynomial complexity on both time and space. Such results are verified by extensive experiments running on our microbenchmark that consists of real-world GPU queries. Furthermore, solutions identified by our method also significantly reduce the total running time of the workload, as compared to sequential and MultiQx-GPU executions.

A GPU-Accelerated Approach for Feature Tracking in Time-Varying Imagery Datasets

We propose a novel parallel connected component labeling (CCL) algorithm along with efficient out-of-core data management to detect and track feature regions of large time-varying imagery datasets. Our approach contributes to the big data field with parallel algorithms tailored for GPU architectures. We remove the data dependency between frames and achieve pixel-level parallelism. Due to the large size, the entire dataset cannot fit into cached memory. Frames have to be streamed through the memory hierarchy (disk to CPU main memory and then to GPU memory), partitioned, and processed as batches, where each batch is small enough to fit into the GPU. To reconnect the feature regions that are separated due to data partitioning, we present a novel batch merging algorithm to extract the region connection information across multiple batches in a parallel fashion. The information is organized in a memory-efficient structure and supports fast indexing on the GPU. Our experiment uses a commodity workstation equipped with a single GPU. The results show that our approach can efficiently process a weather dataset composed of terabytes of time-varying radar images. The advantages of our approach are demonstrated by comparing to the performance of an efficient CPU cluster implementation which is being used by the weather scientists.