Search

In a distributed system, applications can perform both reads and updates without costly synchronous network round-trips by using Conflict-free Replicated Data Types (CRDTs). Most CRDTs are based on some variant of atomic broadcast, as that enables them to support causal dependencies between updates of multiple objects. However, the overhead of this atomic broadcast is unnecessary in systems handling only independent CRDT objects. We identified a set of use cases for tracking resource usage where there is a need for a replication mechanism with less complexity and network usage as compared to using atomic broadcast. In this paper, we present the design of such a replication protocol that efficiently leverages the commutativity of CRDTs. The proposed protocol CReDiT (CRDT enhanced with intelligence) uses up to four communication steps per update, but these steps can be batched as needed. It uses network resources only when updates need to be communicated. Furthermore, it is less sensit ive to server failures than current state-of-the-art solutions as other nodes can use new values already after the first communication step, instead of after two or more.

Embedded systems typically have limited processing and storage capabilities, and may only intermittently be powered on. After sending data from its sensors upstream, the system must therefore be able to trust that the data, once acknowledged, is not lost. The purpose of this work is to propose a novel solution for replicating data between the upstream nodes in such systems, with a minimal effect on the software architecture. On the assumption that there is no relative order between replicated data tuples, we designed a new replication protocol based on partial replication. Our protocol uses only 2 communication steps per data tuple, instead of the 3 to 12 used by other solutions. We verified its failover mechanism in a proof-of-concept implementation of the protocol using simulated network failures, and evaluated the implementation on throughput and latency in several controlled experiments using up to 7 nodes in up to 5 geographically separated areas, with up to 1000 data producers per node. The recorded system throughput increased linearly relative to both the number of nodes and the number of data producers. For comparison, Paxos showed a performance similar to our protocol when using 3 nodes, but got slower as more nodes were added. The lack of a relative order, in combination with partial replication, enables our system to continue working during network partitions, not only in the part containing the majority of the nodes, but also in any sufficiently large minority partitions.

When software communicates, it typically uses some middleware. In this thesis we focus on the group of middleware called “messaging gateways”, which normally uses the store-and-forward architecture. Our overall goal is to find ways to improve the throughput and reliability of such gateways in general, in particular the ones used for forwarding mobile text messages. In short, our work resulted in a new anomaly detection algorithm, a reduction of the number of database operations in a commercial messaging gateway, and two new highly performant data replication protocols. In addressing the throughput, we first wanted a better understanding of how the round-trip times for outgoing requests varied, in order to correctly detect abnormal delays. This resulted in a generalized extension of exponential smoothing, which we used in a novel algorithm to detect anomalies. We then performed an architecture analysis of an industrial messaging gateway based on its quality requirements. From this exploratory case study, we deduced a somewhat unexpected plan to migrate the balance management module to a set of microservices, and identified situations where database operations could be batched in order to also provide higher throughput. A common and easy way to improve the reliability of a system is to replicate the stored data to one or more additional servers. However, we found that this approach often leads to lower throughput due to extra network traffic required. We addressed this by first writing a problem formulation on how the quality attributes of a messaging gateway would be affected by a multi-node configuration, resulting in a review of state of the art and state of practice for multi-node systems. Building on this review, we developed a new data replication algorithm, which we validated in a controlled experiment. Its proof-of-concept implementation showed that even in a geo-distributed configuration, replication throughput can scale with the number of nodes. Text messa

To keep internet based services available despite inevitable local internet and power outages, their data must be replicated to one or more other sites. For most systems using the store-and-forward architecture, data loss can also be prevented by using end-to-end acknowledgements. So far we have not found any sufficiently good solutions for replication of data in store-and-forward systems without acknowledgements and with geographically separated system nodes. We therefore designed a new replication protocol, which could take advantage of the lack of a global order between the messages and the acceptance of a slightly higher risk for duplicated deliveries than existing protocols. We tested a proof-of-concept implementation of the protocol for throughput and latency in a controlled experiment using 7 nodes in 4 geographically separated areas, and observed the throughput increasing superlinearly with the number of nodes up to almost 3500 messages per second. It is also, to the best of our knowledge, the first replication protocol with a bandwidth usage that scales according to the number of nodes allowed to fail and not the total number of nodes in the system.

Background: The Enterprise Messaging Gateway(EMG) from Infoflex Connect (ICAB) is a monolithic system used to deliver mobile text messages (SMS) world-wide. The companies using it have diverse requirements on both functionality and quality attributes and would thus benefit from more versatile customizations, e.g. regarding authorization and data replication. Objective: ICAB needed help in assessing the current architecture of EMG in order to find candidates for architectural changes as well as fulfilling the needs of variability in meeting the wide range of customer requirements. Method: We analysed EMG using a lightweight version of ATAM (Architectural Trade-off Analysis Method) to get a better understanding of how different architectural decisions would affect the trade-offs between the quality requirements from the identified stakeholders. Result: Using the results of this structured approach, it was easy for ICAB to identify the functionality that needed to be improved. It also became clear that the selected component should be converted into a set of microservices, each one optimized for a specific set of customers. Limitation: The stakeholder requirements were gathered intermittently during a long period of continuous engagement, but there is a chance some of their requirements were still not communicated to us. Conclusion: Even though this ATAM study was performed internally at ICAB without direct involvement from any external stakeholders, documenting elicited quality attribute requirements and relating them to the EMG architecture provided new, unexpected, and valuable understandings of the system with a rather small effort.

Mobile text messages (SMS) are sometimes used for authentication, which requires short and reliable delivery times. The observed round-trip times when sending an SMS message provide valuable information on the quality of the connection. In this industry paper, we propose a method for detecting round-trip time anomalies, where the exact distribution is unknown, the variance is several orders of magnitude, and there are lots of shorter spikes that should be ignored. In particular, we show that using an adaption of Double Seasonal Exponential Smoothing to reduce the content dependent variations, followed by the Remedian to find short-term and long-term medians, successfully identifies larger groups of outliers. As training data for our method we use log files from a live SMS gateway. In order to verify the effectiveness of our approach, we utilize simulated data. Our contributions are a description on how to isolate content dependent variations, and the sequence of steps to find significant anomalies in big data.

Businesses often use mobile text messages (SMS) as a cost effective and universal way of communicating concise information to their customers. Today, these messages are usually sent via SMS brokers, which forward them further to the next stakeholder, typically the various mobile operators, and then the messages eventually reach the intended recipients. Infoflex Connect AB delivers an SMS gateway application to the brokers with the main responsibility of reliable message delivery within set quality thresholds. However, the protocols used for SMS communication are not designed for reliability and thus messages may be lost. In this position paper we deduce requirements for a new protocol for routing messages through the SMS gateway application running at a set of broker nodes, in order to increase the reliability. The requirements cover important topics for the required communication protocol such as event ordering, message handling and system membership. The specification of such requirements sets the foundation for the forthcoming design and implementation of such a protocol and its evaluation.

Regression testing research has received significant focus during the past decades, acknowledging the benefits it can provide to organisations in terms of reduced development and maintenance costs, as well as sustained end-user satisfaction. There are several challenges left to overcome before the industry can fully take advantage of the available research results in this area. To get a better overview of how current regression testing research fits in with today’s industrial practices, we read a selection of papers in the field and based on our experience, critically examined their content. As a result, we present and discuss a taxonomy of regression testing challenges, from the perspectives of both methods and organisations, that we believe will foster the industrial uptake of regression testing., TESTMINE - Mining Test Evolution for Improved Software Regression Test Selection (KKS)

To enable fast and reliable delivery of mobile text messages (SMS), special bidirectional protocols are often used. Measuring the achieved throughput and involved latency is however non-trivial, due to the complexity of these protocols. Modifying an existing system would incur too much of a risk, so instead a new tool was created to analyse the log files containing information about this traffic in a black-box fashion. When the produced raw data was converted into graphs, they gave new insights into the behaviour of both the protocols and the remote systems involved., TESTMINE - Mining Test Evolution for Improved Software Regression Test Selection (KKS)