Deep learning (DL) algorithms require, to be fully effective, harvesting an increasingly large amount of data. These data, typically organized as millions of small files, stress filesystems and are difficult to manage. In fact, despite the huge development of DL tools and specialized hardware, data loading pipeline for DL still lacks behind in ease of use, standardization and scalability.In this work we try to rethink the data loading pipeline, by leveraging NoSQL DBs for storing both data and metadata, making them efficiently available through the network, and allowing easier data distribution for parallel DL training. We present our open-source, Apache Cassandra-based data loader and illustrate its use and performance, which enable easy and efficient data management and decentralized data distribution for parallel learning applications.

The adoption of Big Data technologies can potentially boost the scalability of data-driven biology and health workflows by orders of magnitude. Consider, for instance, that technologies in the Hadoop ecosystem have been successfully used in data-driven industry to scale their processes to levels much larger than any biological-or health-driven work attempted thus far. In this work we demonstrate the scalability of a sequence alignment pipeline based on technologies from the Hadoop ecosystem - namely, Apache Flink and Hadoop MapReduce, both running on the distributed Apache YARN platform. Unlike previous work, our pipeline starts processing directly from the raw BCL data produced by Illumina sequencers. A Flink-based distributed algorithm reconstructs reads from the Illumina BCL data, and then demultiplexes them - analogously to the bcl2fastq2 program provided by Illumina. Subsequently, the BWA-MEM-based distributed aligner from the Seal project is used to perform read mapping on the YARN platform. While the standard programs by Illumina and BWA-MEM are limited to shared-memory parallelism (multi-threading), our solution is completely distributed and can scale across a large number of computing nodes. Results show excellent pipeline scalability, linear in the number of nodes. In addition, this approach automatically benefits from the robustness to hardware failure and transient cluster problems provided by the YARN platform, as well as the scalability of the Hadoop Distributed File System. Moreover, this YARN-based approach complements the up-and-coming version 4 of the GATK toolkit, which is based on Spark and therefore can run on YARN. Together, they can be used to form a scalable complete YARN-based variant calling pipeline for Illumina data, which will be further improved with the arrival of distributed in-memory filesystem technology such as Apache Arrow, thus removing the need to write intermediate data to disk.

We present three lock-free data structures for priority task scheduling: a priority work-stealing one, a centralized one with ρ-relaxed semantics, and a hybrid one combining both concepts. With the single-source shortest path (SSSP) problem as example, we show how the different approaches affect the prioritization and provide upper bounds on the number of examined nodes. We argue that priority task scheduling allows for an intuitive and easy way to parallelize the SSSP problem, notoriously a hard task. Experimental evidence supports the good scalability of the resulting algorithm. The larger aim of this work is to understand the trade-offs between scalability and priority guarantees in task scheduling systems. We show that ρ-relaxation is a valuable technique for improving the first, while still allowing semantic constraints to be satisfied: the lock-free, hybrid $k$-priority data structure can scale as well as work-stealing, while still providing strong priority scheduling guarantees, which depend on the parameter k. Our theoretical results open up possibilities for even more scalable data structures by adopting a weaker form of ρ-relaxation, which still enables the semantic constraints to be respected.

We present a new, deadlock-free, routing scheme for toroidal interconnection networks, called OutFlank Routing (OFR). OFR is an adaptive strategy which exploits non-minimal links, both in the source and in the destination nodes. When minimal links are congested, OFR deroutes packets to carefully chosen intermediate destinations, in order to obtain travel paths which are only an additive constant longer than the shortest ones. Since routing performance is very sensitive to changes in the traffic model or in the router parameters, an accurate discrete-event simulator of the toroidal network has been developed to empirically validate OFR, by comparing it against other relevant routing strategies, over a range of typical real-world traffic patterns. On the 16x16x16 (4096 nodes) simulated network OFR exhibits improvements of the maximum sustained throughput between 14% and 114%, with respect to Adaptive Bubble Routing.

Optimistic parallelization is a promising approach for the parallelization of irregular algorithms: potentially interfering tasks are launched dynamically, and the runtime system detects conflicts between concurrent activities, aborting and rolling back conflicting tasks. However, parallelism in irregular algorithms can be a function of input parameters, and the amount of parallelism can vary dramatically during the execution. Therefore, determine how many processors should be allocated to execute (the processor allocation problem) for irregular algorithms is very difficult. In this work, we outline the first systematic strategy for addressing this problem.

This paper deals with the application of hybrid automatic repeat request (H-ARQ) techniques to reliable data communications in wireless 3G networks, whose typical applications have to match accurate quality of service (QoS) requirements. A common approach to guarantee such services enhances the H-ARQ features by exploiting turbo codes error correction capabilities. This paper, following the previous approach, investigates the further benefits achievable by developing a strategy that establishes a connection between packet retransmissions and decoding results. Two kinds of these techniques, belonging to soft recombining schemes, are proposed and applied, respectively, to consecutive received packet replicas, or to decoding algorithm outputs, in both cases, without remarkable hardware complexity increase. By means of analytical derivations and simulations, focused on a typical universal mobile telecommunication system (UMTS) environment, a noticeable bit error rate (BER) and, then, frame error rate (FER) improvements have been shown. For the proposed schemes, whenever communications suffer from low signal to noise ratio (SNR) values, significant link throughput increase is also pointed out. As a consequence, these techniques can assure a lower packet delivery delay, besides avoiding transmitted power wasting, if compared with protocols commonly used in wired networks.

This paper deals with the application of hybrid automatic repeat request (H-ARQ) techniques to reliable data communications in wireless 3G networks. Basically, retransmission of coded data is endowed with soft combining schemes applied, respectively, to packet replicas, or to decoding algorithm outputs. By means of analytical derivations and simulations on UMTS environment, a noticeable bit error rate (BER) and, then, frame error rate (FER) improvements have been shown. A significant link throughput increase is also pointed out with a consequent reduced packet delivery delay and power wasting.