Thomas Fuhrmann

Consistency in hindsight: A fully decentralized STM algorithm

Software transactional memory (STM) algorithms often rely on centralized components to achieve atomicity, isolation and consistency. In a distributed setting, centralized components are undesirable as they impair scalability. This paper presents Decent STM, a fully decentralized object-based STM algorithm. It relies on mostly immutable data structures, which are well-suited for replication and migration. It is the first decentralized STM implementing snapshot isolation semantics. A novel randomized consensus protocol guarantees consistency of the mutable parts. Transactions may proceed tentatively before consensus has been reached. Object versioning ensures consistency in hindsight. Thus, atomic code sections never block during execution. The evaluation of benchmarks shows that the guaranteed success of reads more than compensates for the higher conflict rate during commit.

IgorFs: A Distributed P2P File System

IgorFs is a distributed, decentralized peer-to-peer (P2P) file system that is completely transparent to the user. It is built on top of the Igor peer-to-peer overlay network, which is similar to Chord, but provides additional features like service orientation or proximity neighbor and route selection. IgorFs offers an efficient means to publish data files that are subject to frequent but minor modifications. In our demonstration we show two use cases for IgorFs: the first example is (static) software-distribution and the second example is (dynamic) file distribution.

The architecture of the DecentVM: towards a decentralized virtual machine for many-core computing

Fully decentralized systems avoid bottlenecks and single points of failure. Thus, they can provide excellent scalability and very robust operation. The DecentVM is a fully decentralized, distributed virtual machine. Its simplified instruction set allows for a small VM code footprint. Its partitioned global address space (PGAS) memory model helps to easily create a single system image (SSI) across many processors and processor cores. Originally, the VM was designed for networks of embedded 8-bit processors. Meanwhile, it also aims at clusters of many core processors. This paper gives a brief overview of the DecentVM architecture.

Using Linearization for Global Consistency in SSR

Novel routing algorithms such as scalable source routing (SSR) and virtual ring routing (VRR) need to set up and maintain a virtual ring structure among all the nodes in the network. The iterative successor pointer rewiring protocol (ISPRP) is one way to bootstrap such a network. Like its VRR-analogon, ISPRP requires one of the nodes to flood the network to guarantee consistency. Recent results on self-stabilizing algorithms now suggest a new approach to bootstrap the virtual rings of SSR and VRR. This so-called linearization method does not require any flooding at all. Moreover, it has been shown that linearization with shortcut neighbors has on average polylogarithmic convergence time, only.

Experimental Study on Staged Injection in a Supersonic Combustor

At the supersonic combustion test facility of the Institute of Flight Propulsion of the Technische Universitaet Muenchen experimental investigations on staged injection were conducted. In the model combustor a central strut injection system which was inves tigated at the Institute for the last years served as the first stage. As second stage wall injectors with a single injection hole were used. First, an investigation concerning the ignition and flame stabilization capabilities of three different strut inje ctors was conducted. In addition one injector was studied in more detail in combination with the second stage. By means of the measured pressure distribution along the combustor wall and visual observation the ignition of the second stage, the flame stabil ization and the interaction between the two stages was examined depending on different equivalence ratios. Furthermore, the results of the experiments were post processed using an one dimensional preliminary design tool.

Application of DHT-Inspired Routing for Object Tracking

A major problem in tracking objects in sensor networks is trading off update traffic and timeliness of the data that is available to a monitoring site. Typically, either all objects regularly update some central registry with their location information, or the monitoring instance floods the network with a request when it needs information for a particular object. More sophisticated approaches use a P2P-like distributed storage structure on top of geographic routing. The applicability of the latter is limited to certain topologies, and having separate storage and routing algorithms reduces efficiency. In this paper, we present a different solution which is based on the scalable source routing (SSR) protocol. SSR is a network layer routing protocol that has been inspired by distributed hash tables (DHT). It provides key-based routing in large networks of resource-limited devices such as sensor networks. We argue that this approach is more suitable for object tracking in sensor networks because it evenly spreads the updates over the whole network without being limited to a particular network topology. We support our argument with extensive simulations.

Lifting the barriers --- reducing latencies with transparent transactional memory

Synchronization in distributed systems is expensive because, in general, threads must stall to obtain a lock or to operate on volatile data. Transactional memory, on the other hand, allows speculative execution so that it can hide the latencies that are inherent to distributed systems. In this paper, we discuss how transactional memory can carry over to code that uses Javas synchronization means i.e. monitors and volatiles. We show that we can guarantee correct execution according to the Java memory model (JMM) without having to stall at synchronization points. To this end, we use a multi-version software transactional memory system that executes JMM synchronization operations asynchronously. If any such execution has violated the JMM, the transaction rolls back. As a result, only blocking operations require immediate synchronization barriers.

Here is your peer! — Locating peers on a regional level with network coordinates

It seems to be common wisdom that peer-to-peer overlay networks should reflect the structure of the underlying network to minimize the application-level latency, to increase the throughput, and to optimize the traffic for the Internet service providers. Network coordinates are one means to achieve this goal. However, previous real-world experiments were disappointing because methods like Vivaldi could not resolve the structure of the underlying network better than at a continental level. In our work, we have closely analyzed the reasons for that disappointing performance. Thereby, we were able to improve Vivaldi-like network coordinate systems so that they can now differentiate peers at a regional level. In this paper, we report on a large measurement study that we performed with the Vuze BitTorrent client. We show that these improved algorithms can indeed differentiate among different autonomous systems and different geographic regions. We also show that this achievement is less important than commonly thought and that the key contribution lies rather in the latency-optimal overlay formation itself.

Using asymmetric links to improve SSR's routing performance

Empirical evidence shows that asymmetric links occur frequently in wireless networks. Still, most routing protocols assume symmetric links. Thus, they cannot exploit the additional connectivity and bandwidth that asymmetric links provide. In this paper, we propose an extension of the scalable source routing protocol (SSR) that enables SSR to use asymmetric links. It enhances SSRs hello messages with a limited amount of local topology information, which allows the nodes to discover directed cycles in the network. Such cycles provide a return path in cases where the forward path contains asymmetric links. With our proposed extension, SSR can thus use asymmetric links and still guarantee bidirectional connectivity. We evaluate our proposals with extensive simulations. A practical test in a 150 node test bed is ongoing work.

Using link-layer broadcast to improve scalable source routing

Scalable source routing (SSR) is a network layer routing protocol that provides services that are similar to those of structured peer-to-peer overlays. In this paper, we describe several improvements to the SSR protocol. They aim at providing nodes with more up-to-date routing information: 1. The use of link-layer broadcast enables all neighbors of a node to contribute to the forwarding process. 2. A light-weight and fast selection mechanism avoids packet duplication and optimizes the source route iteratively. 3. Nodes implicitly learn the networks topology from overheard broadcast messages. We present simulation results which show the performance gain of the proposed improvements: 1. The delivery ratio in settings with high mobility increases. 2. The required per-node state can be reduced as compared with the original SSR protocol. 3. The route stretch decreases. --- These improvements are achieved without increasing the routing overhead.