Search

High-speed data transfer is vital to data-intensive computing that often requires moving large data volumes efficiently within a local data center and among geographically dispersed facilities. Effective utilization of the abundant resources in modern multicore environments for data transfer remains a persistent challenge, particularly, for Non-Uniform Memory Access (NUMA) systems wherein the locality of data accessing is an important factor. This requires rethinking how to exploit parallel access to data and to optimize the storage and network I/Os. We address this challenge and present a novel design of asynchronous processing and resource-aware task scheduling in the context of high-throughput data replication. Our software allocates multiple sets of threads to different stages of the processing pipeline, including storage I/O and network communication, based on their capacities. Threads belonging to each stage follow an asynchronous model, and attain high performance via multiple locality-aware and peer-aware mechanisms, such as task grouping, buffer sharing, affinity control and communication protocols. Our design also integrates high performance features to enhance the scalability of data transfer in several scenarios, e.g., file-level sorting, block-level asynchrony, and thread-level pipelining. Our experiments confirm the advantages of our software under different types of workloads and dynamic environments with contention for shared resources, including a 28-160 percent increase in bandwidth for transferring large files, 1.7-66 times speed-up for small files, and up to 108 percent larger throughput for mixed workloads compared with three state of the art alternatives, GridFTP , BBCP and Aspera .

Wide area data transfer may be a major bottleneck for the end-to-end performance of distributed applications. A practical way of increasing the wide area throughput at the application layer is using multiple parallel streams. Although increased number of parallel streams may yield much better performance than using a single stream, overwhelming the network by opening too many streams may have an inverse effect. The congestion created by excess number of streams may cause a drop down in the throughput achieved. Hence, it is important to decide on the optimal number of streams without congesting the network. Predicting this "optimum” number is not straightforward, since it depends on many parameters specific to each individual transfer. Generic models that try to predict this number either rely too much on historical information or fail to achieve accurate predictions. In this paper, we present a set of new models which aim to approximate the optimal number with least history information and lowest prediction overhead. An algorithm is introduced to select the best combination of historic information to do the prediction for evaluation purposes as well as optimizing prediction by reducing error rate. We measure the feasibility and accuracy of the proposed prediction models by comparing to actual GridFTP data transfer by using little historical information and have seen that we could predict the throughput of parallel streams accurately and find a very close approximation of the optimal stream number.

In this paper, we present the design and implementation of an application-layer data throughput prediction and optimization service for many-task computing in widely distributed environments. This service uses multiple parallel TCP streams to improve the end-to-end throughput of data transfers. A novel mathematical model is developed to determine the number of parallel streams, required to achieve the best network performance. This model can predict the optimal number of parallel streams with as few as three prediction points. We implement this new service in the Stork Data Scheduler, where the prediction points can be obtained using Iperf and GridFTP samplings. Our results show that the prediction cost plus the optimized transfer time is much less than the nonoptimized transfer time in most cases. As a result, Stork data transfer jobs with optimization service can be completed much earlier, compared to nonoptimized data transfer jobs.

This paper explores the feasibility of a storage architecture that offers the reliability and access performance characteristics of a high-end system, yet is cost-efficient. We propose ThriftStore, a storage architecture that integrates two types of components: volatile, aggregated storage and dedicated, yet low-bandwidth durable storage. On the one hand, the durable storage forms a back end that enables the system to restore the data the volatile nodes may lose. On the other hand, the volatile nodes provide a high-throughput front-end. Although integrating these components has the potential to offer a unique combination of high throughput and durability at a low cost, a number of concerns need to be addressed to architect and correctly provision the system. To this end, we develop analytical and simulation-based tools to evaluate the impact of system characteristics (e.g., bandwidth limitations on the durable and the volatile nodes) and design choices (e.g., the replica placement scheme) on data availability and the associated system costs (e.g., maintenance traffic). Moreover, to demonstrate the high-throughput properties of the proposed architecture, we prototype a GridFTP server based on ThriftStore. Our evaluation demonstrates an impressive, up to 800 Mbps transfer throughput for the new GridFTP service.