André Schumacher

SeqPig: simple and scalable scripting for large sequencing data sets in Hadoop

Summary: Hadoop MapReduce-based approaches have become increasingly popular due to their scalability in processing large sequencing datasets. However, as these methods typically require in-depth expertise in Hadoop and Java, they are still out of reach of many bioinformaticians. To solve this problem, we have created SeqPig, a library and a collection of tools to manipulate, analyze and query sequencing datasets in a scalable and simple manner. SeqPigscripts use the Hadoop-based distributed scripting engine Apache Pig, which automatically parallelizes and distributes data processing tasks. We demonstrate SeqPig’s scalability over many computing nodes and illustrate its use with example scripts. Availability and Implementation: Available under the open source MIT license at http://sourceforge.net/projects/seqpig/ Contact: andre.schumacher@yahoo.com Supplementary information: Supplementary data are available at Bioinformatics online.

Online bin packing with delay and holding costs

We consider online bin packing where in addition to the opening cost of each bin, the arriving items collect delay costs until their assigned bins are released (closed), and the open bins themselves collect holding costs. Besides being of practical interest, this problem generalises several previously unrelated online optimisation problems. We provide a general online algorithm for this problem with competitive ratio 7, with improvements for the special cases of zero delay or holding costs and size-proportional item delay costs.

Balanced Multipath Source Routing

We consider the problem of balancing the traffic load ideally over a wireless multihop network. In previous work, a systematic approach to this task was undertaken, starting with an approximate optimisation method that guarantees a provable congestion performance bound, and then designing a distributed implementation by modifying the DSR protocol. In this paper, the performance of the resulting Balanced Multipath Source Routing ( BMSR ) protocol is validated in a number of simulated networking scenarios. In particular, we study the effect of irregular network structure on the performance of the protocol, and compare it to the performance of DSR and an idealised shortest-path routing algorithm in setups with several source-destination pairs. For all network scenarios we consider, BMSR outperforms DSR significantly. BMSR is also shown to be more robust than the shortest-path algorithm, in that it can distribute the traffic load more evenly in cases where shortest-path routing is impeded by radio interference between proximate paths.

Distributed computation of maximum lifetime spanning subgraphs in sensor networks

We present a simple and efficient distributed method for determining the transmission power assignment that maximises the lifetime of a data-gathering wireless sensor network with stationary nodes and static power assignments. Our algorithm determines the transmission power level inducing the maximum-lifetime spanning subgraph of a network by means of a distributed breadth-first search for minmax-power communication paths, i.e. paths that connect a given reference node to each of the other nodes so that the maximum transmission power required on any link of the path is minimised. The performance of the resulting Maximum Lifetime Spanner (MLS) protocol is validated in a number of simulated networking scenarios. In particular, we study the performance of the protocol in terms of the number of required control messages, and compare it to the performance of a recently proposed Distributed Min-Max Tree (DMMT) algorithm. For all network scenarios we consider, MLS outperforms DMMT significantly. We also discuss bringing down the message complexity of our algorithm by initialising it with the Relative Neighbourhood Graph (RNG) of a transmission graph rather than the full graph, and present an efficient distributed method for reducing a given transmission graph to its RNG.

Load balancing by distributed optimisation in ad hoc networks

We approach the problem of load balancing for wireless multi-hop networks by distributed optimisation. We implement an approximation algorithm for minimising the maximum network congestion as a modification to the DSR routing protocol. The algorithm is based on shortest-path computations that are integrated into the DSR route discovery and maintenance process. The resulting Balanced Multipath Source Routing (BMSR) protocol does not need to disseminate global information throughout the network. Our simulations with the ns2 simulator show a gain of 14% to 69% in the throughput, depending on the setup, compared to DSR for a high network load.

Distributed algorithms for lifetime maximization in sensor networks via Min---Max spanning subgraphs

We consider the problem of static transmission-power assignment for lifetime maximization of a wireless sensor network with stationary nodes operating in a data-gathering scenario. Using a graph-theoretic approach, we propose two distributed algorithms, MLS and BSpan, that construct spanning trees with minimum maximum (minmax) edge cost. MLS is based on computation of minmax-cost paths from a reference node, while BSpan performs a binary search over the range of power levels and exploits the wireless broadcast advantage. We also present a simple distributed method for pruning a graph to its Relative Neighborhood Graph, which reduces the worst-case message complexity of MLS under natural assumptions on the path-loss. In our network simulations both MLS and BSpan significantly outperform the recently proposed Distributed Min–Max Tree algorithm in terms of number of messages required.

Distributed Sleep Scheduling in Wireless Sensor Networks via Fractional Domatic Partitioning

We consider setting up sleep scheduling in sensor networks. We formulate the problem as an instance of the fractional domatic partition problem and obtain a distributed approximation algorithm by applying linear programming approximation techniques. Our algorithm is an application of the Garg-Konemann (GK) scheme that requires solving an instance of the minimum weight dominating set (MWDS) problem as a subroutine. Our two main contributions are a distributed implementation of the GK scheme for the sleep-scheduling problem and a novel asynchronous distributed algorithm for approximating MWDS based on a primal-dual analysis of Chvatals set-cover algorithm. We evaluate our algorithm with ns2 simulations.

Approximately Uniform Online Checkpointing with Bounded Memory

In many complex computational processes one may want to store a sample of the process’ history for later use by placing checkpoints. In this paper we consider the problem of maintaining, in an online fashion, a collection of k checkpoints as an approximately uniformly spaced sample in the history of a continuous-time process. We present deterministic algorithms tailored for small values of k and a general one for arbitrary k. The algorithms are proven to be close to optimum for several different measures.

Distributed network utility maximization in wireless networks with a bounded number of paths

We consider the fair multicommodity flow problem in multihop wireless networks. We optimize the flow between a number of source-destination pairs to achieve a fair allocation of network resources while satisfying node capacity constraints. To account for the limitations of wireless devices, we use a path-based formulation and allow only a bounded number of paths between each source-destination pair. We develop a dual decomposition based distributed algorithm for solving the problem, show the optimality of a stationary solution, and compare the performance of the algorithm with a centralized branch-and-bound method.

On the power of lookahead in online lot-sizing

We propose an online algorithm for an economic lot-sizing (ELS) problem with lookahead, which achieves asymptotically optimal worst-case performance for increasing lookahead. Although intuitive, this result is interesting since deterministic algorithms for previously studied online ELS problems have unbounded competitive ratio. We also prove lookahead-dependent lower bounds for deterministic algorithms.