Luís E. T. Rodrigues

Distributed Systems for System Architects

The distributed systems architect assembles pieces of hardware that are at least as large as a computer or a network router, and assigns pieces of software that are self-contained - such as Java applets - to those hardware components. As system complexity, size and diversity grow, the probability of inconsistency, unreliability, non-responsiveness and insecurity, increases. It is absolutely necessary for distributed systems architects to understand the management of such complex systems. Distributed Systems for System Architects addresses these issues.

Appia, a flexible protocol kernel supporting multiple coordinated channels

Distributed applications are becoming increasingly complex, often requiring the simultaneous use of several communication channels with different qualities-of-service. This paper presents the Appia system, a protocol kernel that supports applications requiring multiple coordinated channels. Appia offers a clean and elegant way for the application to express inter-channel constraints, such as, for instance, that all channels should provide consistent information about the failures of remote nodes. These constraints can be implemented as protocol layers that can be dynamically combined with other protocol layers.

Piecewise-affine state feedback for piecewise-affine slab systems using convex optimization

Abstract This paper shows that Lyapunov-based state feedback controller synthesis for piecewise-affine (PWA) slab systems can be cast as an optimization problem subject to a set of linear matrix inequalities (LMIs) analytically parameterized by a vector. Furthermore, it is shown that continuity of the control inputs at the switchings can be guaranteed by adding equality constraints to the problem without affecting its parameterization structure. Finally, it is shown that piecewise-affine state feedback controller synthesis for piecewise-affine slab systems to maximize the decay rate of a quadratic control Lyapunov function can be cast as a set of quasi-concave optimization problems analytically parameterized by a vector. Before casting the synthesis in the format presented in this paper, Lyapunov-based piecewise-affine state feedback controller synthesis could only be formulated as a bi-convex optimization program, which is very expensive to solve globally. Thus, the fundamental importance of the contributions of the paper relies on the fact that, for the first time, the piecewise-affine state feedback synthesis problem has been formulated as a convex problem with a parameterized set of LMIs that can be relaxed to a finite set of LMIs and solved efficiently to a point near the global optimum using available software. Furthermore, it is shown for the first time that, in some situations, the global can be exactly found by solving only one concave problem.

D2STM: Dependable Distributed Software Transactional Memory

At current date the problem of how to build distributed and replicated Software Transactional Memory (STM) to enhance both dependability and performance is still largely unexplored. This paper fills this gap by presenting D2STM, a replicated STM whose consistency is ensured in a transparent manner, even in the presence of failures. Strong consistency is enforced at transaction commit time by a non-blocking distributed certification scheme, which we name BFC (Bloom Filter Certification). BFC exploits a novel Bloom Filter-based encoding mechanism that permits to significantly reduce the overheads of replica coordination at the cost of a user tunable increase in the probability of transaction abort. Through an extensive experimental study based on standard STM benchmarks we show that the BFC scheme permits to achieve remarkable performance gains even for negligible (e.g. 1%) increases of the transaction abort rate.

CesiumSpray>: a Precise and Accurate Global Time Servicefor Large-scale Systems

In large-scale systems, such as Internet-based distributed systems, classical clock-synchronization solutions become impractical or poorly performing, due to the number of nodes and/or the distance among them. We present a global time service for world-wide systems, based on an innovative clock synchronization scheme, named CesiumSpray. The service exhibits high precision and accuracy; it is virtually indefinitely scalable; and it is fault-tolerant. It is deterministic for real-time machinery in the local area, which makes it particularly well-suited for, though not limited to, large-scale real-time systems. The main features of our clock synchronization scheme can be summarized as follows: hybrid external/internal synchronization protocol improves effectiveness of synchronization; heterogeneous failure semantics for clocks and processors improves previous lower bounds on processors; two-level hierarchy improves scalability. The root of the hierarchy is the GPS satellite constellation, which “sprays” its reference time over a set of nodes provided with GPS receivers, one per local network. The second level of the hierarchy performs internal synchronization, further “spraying” the external time inside the local network.

Observer-based control of piecewise-affine systems

This paper presents a new synthesis method for both state and dynamic output feedback control of a class of hybrid systems called piecewise-affine (PWA) systems. The synthesis procedure delivers stabilizing controllers that can be proven to give either asymptotic or exponential convergence rates. The synthesis method builds on existing PWA stability analysis tools by transforming the design into a closed-loop analysis problem wherein the controller parameters are unknown. More specifically, the proposed technique formulates the search for a piecewise-quadratic control Lyapunov function and a piecewise-affine control law as an optimization problem subject to linear constraints and a bilinear matrix inequality. The linear constraints in the synthesis guarantee that sliding modes are not generated at the switching. The resulting optimization problem is known to be hard, but suboptimal solutions can be obtained using the three iterative algorithms presented in the paper. The new synthesis technique allows controllers to be designed with a specified structure, such as a combined regulator and observer. The observers in these controllers then enable switching based on state estimates rather than on measured outputs. The overall design approach, including a comparison of the synthesis algorithms and the performance of the resulting controllers, is clearly demonstrated in four simulation examples.

Fault-tolerant broadcasts in CAN

Fault-tolerant distributed systems based on field-buses may take advantage from reliable and atomic broadcast. There is a current belief that CAN native mechanisms provide atomic broadcast. In this paper, we dismiss this misconception, explaining how network errors may lead to: inconsistent message delivery; generation of message duplicates. These errors may occur when faults hit the last two bits of the end of frame delimiter. Although rare, its influence cannot be ignored, for highly fault-tolerant systems. Finally, we give a protocol suite that handles the problem effectively.

The Delta-4 extra performance architecture (XPA)

The design of an extra performance architecture for Delta-4, which explicitly supports the requirements of real-time systems with respect to throughput and response, is presented. The Delta-4 approach to fault tolerance is based on the replication of software components on distinct host computers using a range of different replication strategies. The problems of replicate divergence are discussed, and a solution based on message selection and preemption synchronization messages is proposed. A description of the ongoing implementation of such a system within the overall Delta-4 framework is included.<<ETX>>

AMp: a highly parallel atomic multicast protocol

This paper deals with the problem of reliable group communication for distributed applications, in the context of the Reliable Broadcast class of protocols. An atomic multicast protocol for token passing Lans is presented. The actual implementation is on an 8802/4 Token-bus, although it is applicable to 8802/5 Token-rings and the FDDI Fibre-Optic network. The simplicity and efficiency of reliable broadcast protocols may be considerably improved, if the system fault model is restricted or convenient architectures are used. Fail-controlled communication components are used here to build an efficient reliable multicast protocol on top of the exposed MAC interface of a VLSI Lan controller. The architecture is built on standard Lans, in view of taking advantage of the availability of communications hardware and the possibility of coexistence with standard stations, in the same network. The service offered allows transparent multicasting inside logical groups, which are dynamically created and updated. The primitive is highly parallel and provides atomic agreement and consistent delivery order, respecting logical precedence. These features are an important contribution for the implementation of high performance distributed computing systems.

When Scalability Meets Consistency: Genuine Multiversion Update-Serializable Partial Data Replication

In this article we introduce GMU, a genuine partial replication protocol for transactional systems, which exploits an innovative, highly scalable, distributed multiversioning scheme. Unlike existing multiversion-based solutions, GMU does not rely on a global logical clock, which represents a contention point and can limit system scalability. Also, GMU never aborts read-only transactions and spares them from distributed validation schemes. This makes GMU particularly efficient in presence of read-intensive workloads, as typical of a wide range of real-world applications. GMU guarantees the Extended Update Serializability (EUS) isolation level. This consistency criterion is particularly attractive as it is sufficiently strong to ensure correctness even for very demanding applications (such as TPC-C), but is also weak enough to allow efficient and scalable implementations, such as GMU. Further, unlike several relaxed consistency models proposed in literature, EUS has simple and intuitive semantics, thus being an attractive, scalable consistency model for ordinary programmers. We integrated the GMU protocol in a popular open source in-memory transactional data grid, namely Infinispan. On the basis of a large scale experimental study performed on heterogeneous experimental platforms and using industry standard benchmarks (namely TPC-C and YCSB), we show that GMU achieves linear scalability and that it introduces negligible overheads (less than 10%), with respect to solutions ensuring non-serializable semantics, in a wide range of workloads.