Alexander Stanik

Detecting bottlenecks in parallel DAG-based data flow programs

In recent years, several frameworks have been introduced to facilitate massively-parallel data processing on shared-nothing architectures like compute clouds. While these frameworks generally offer good support in terms of task deployment and fault tolerance, they only provide poor assistance in finding reasonable degrees of parallelization for the tasks to be executed. However, as billing models of clouds enable the utilization of many resources for a short period of time for the same cost as utilizing few resources for a long time, proper levels of parallelization are crucial to achieve short processing times while maintaining good resource utilization and therefore good cost efficiency. In this paper, we present and evaluate a solution for detecting CPU and I/O bottlenecks in parallel DAG-based data flow programs assuming capacity constrained communication channels. The detection of bottlenecks represents an important foundation for manually or automatically scaling out and tuning parallel data flow programs in order to increase performance and cost efficiency.

Hardware as a Service (HaaS): Physical and virtual hardware on demand

Cloud computing has already been adopted in a broad range of application domains. However, domains like the distributed development of embedded systems are still unable to benefit from the advancements of cloud computing. Besides general security concerns, a common obstacle often is the incompatibility between such applications and the cloud. In particular, if applications need direct access to hardware elements, cloud computing cannot be used. In this paper we describe an approach of a novel cloud layer called Hardware as a Service (HaaS), which allows for usage distinct hardware components through the Internet analogously to the cloud services. HaaS focuses the transparent integration of remote hardware that is distributed over multiple geographical locations into an operating system. Furthermore, HaaS will not only enable interconnection of physical systems, but also virtual hardware emulation. Therefore, we consider in this paper only the use of emulated hardware and the interconnection with hardware models. To demonstrate the tremendous improvement by a Haas cloud, we explain the applicability in a distributed development process by an anti blocking system and an adaptive cruise control system in the automotive industry.

SLA-driven Federated Cloud Networking: Quality of Service for Cloud-based Software Defined Networks

Abstract Since the cloud paradigm becomes more and more popular for the dynamic resources allocation, new techniques and performance improvements for scalability as well as new cloud services on all three layers of the cloud stack were developed. Furthermore, another well covered topic is cloud federation concerning processing power and strength. However, the flexibility of the cloud is limited in terms of network services and federated networking between autonomous cloud data-center. With the promising opportunities of Software Defined Networking (SDN), this gap can be closed and enables cloud environments to establish networking federations between autonomous data-centers and virtual network partitioning in a single cloud infrastructure. In this paper we are introducing and describing an architectural approach for a generic layered model and API based software architecture to orchestrate and federate heterogeneous networks. In particular, we present an architecture that enables Quality of Service (QoS) aware configurations of network resources in a cloud infrastructure of one data-center and federated networking between different SDN based cloud networks over and above the data-center network edge. Furthermore, this architecture uses a Service Level Agreement (SLA) protocol and language to expose Key Performance Indicators (KPI) and to negotiate appropriated QoS constrains which are applied to the virtually sliced underlying network substrate. In this way, capabilities of the orchestration and the current utilization of the network are building the foundation for dynamic negotiated SLAs and the within guaranteed QoS of network resources. The approach presented in this paper will change todays IT landscape and allows every organization to purchase required network characteristics on demand.

Service-level agreement aggregation for quality of service-aware federated cloud networking

Since the cloud paradigm becomes increasingly popular for dynamic resources allocation, the flexibility of a cloud is still limited regarding network services and their autonomous federation between different providers. The following architectural approach introduces a generic layered model to orchestrate and federate heterogeneous networks. In particular, an architecture is presented that enables quality of service (QoS) aware parameterisation of network resources in a cloud infrastructure of a single data-centre as well as for a federation. Furthermore, this architecture uses a service-level agreement (SLA) protocol and language to expose key performance indicators and to negotiate appropriate QoS constrains that are applied to the virtually sliced underlying network substrate. In this way, capabilities of the orchestration and the current utilisation of the network are building the foundation for dynamic negotiated SLAs and the within-guaranteed QoS network resources. Therefore an aggregation mechanism is illustrated for merging service-level objectives and for guaranteeing a single SLA that specifies obligations and responsibilities of all participants.

Portable SDN Applications on the PaaS Layer

SaaS applications often face a vendor or technical lock-in due to PaaS provider specific specifications, like cloud management APIs. As a solution, this paper presents a novel approach for developing applications more PaaS provider independent. In particular, the approach illustrates advantages of extending JavaEE application servers with a new container that executes so called Infrastructure Java Beans and provides a new set of state-of-the-art cloud APIs, which for instance enable Software Defined Networking capabilities. This container allows to manage and observe the application server itself as well as the cloud infrastructure, including compute and network resources. Thus, the application server can scale on demand controlled by the application itself and adjusted with optimized network functions that are designed application specific. In addition, all of this advantages are invested with a cloud provider independent migration possibility.

An approach for QoS constraint networks in cloud environments

Many QoS constraints in network environments exists, beginning with hard real time networks on the field bus level, up to Differentiated Services or VLAN tagging for commodity computer networks. However, all these approaches are not applicable for cloud-based environments, where networks need to be configured dynamically and on-demand. In this paper, we present an approach for centralized and dynamic network configuration. We show that QoS constraints can be fulfilled by decoupling the control and datapath of packet forwarding hardware in Ethernet based interconnection networks, that are widely-used in data centers and cloud environments. Furthermore, we introduce our cloud middleware and explain the resulting problems raised by interconnecting virtualized embedded systems through overlay networks with QoS constraints coming from the field bus domain.

Failover Pattern with a Self-Healing Mechanism for High Availability Cloud Solutions

Cloud computing has already been adopted in a broad range of application domains and has become an established building block in IT landscapes. During the process of cloud middleware development, the companies have focused mainly on the high availability of data and end-user services, but unfortunately neglected the availability of middleware components. Therefore failures of the middleware components itself usually leads to a partial or even total blackout of the cloud. In this paper, we present the design and implementation of a novel scalable and highly available multi-master pattern for cloud middlewares. In contrast to existing Infrastructure-as-a-Service cloud management frameworks, which are usually designed in a centralized tree topology composed in a three-tiered master worker architecture, we introduce a concept for a multi tree with all tree roots connected in a fully connected mesh topology. In this architecture user requests are load balanced over multiple failover servers. Furthermore, our concept includes an automatic self-healing mechanism for worker nodes of each tree.

Scalable decentralised data fusion using hypercube gossiping

We present a scalable approach to distributed data fusion based on the bandwidth- and latency-efficient hypercube topology. Provided the individual estimation errors are uncor-related, double-counting is avoided altogether by exploitation of the network topology, without the need for approximation with channel filters or pedigree logs. The fully allocated network is symmetric and exhibits no special nodes or network locations. Our approach uses simple O (log n) systolic information dissemination without the need to first determine an ad-hoc communication scheme. As a secondary contribution we present practical considerations on how to reduce bandwidth for severely limited bandwidth scenarios and present an optimised timing mode that reduces mean information latency.

A proposal for REST with XMPP as base protocol for intercloud communication

The IEEE Intercloud project aims to facilitate intercloud interoperability and portability. While topology elements and basic security and trust models had been developed during the last years, an adequate base protocol for the intercloud communication is still required. This protocol have to be extensible and should eliminate limitations of HTTP. This paper proposes a novel protocol that addresses the requirements of the P2302 standard to design a service framework that provides similar capabilities as HTTP-based SOAP or REST web services. In fact, the protocol proposed in this paper adopts the REST architectural style to XMPP. Furthermore, the protocol allows for advertising resource capabilities, discovering resources, and to perform actions on resources by its nature. This protocol has been implemented in the RWX4J framework and was evaluated regarding functionality, performance, and efficiency. The results presented in this paper show significant improvements compared to HTTP-based RESTful web services.

MO-BIZZ: Fostering Mobile Business through Enhanced Cloud Solutions

Mobile devices are becoming more powerful and are able to support applications of significant size and high processing efforts. However, most of the applications require mobile internet connections to use information and services of background infrastructures. Especially business applications for mobile devices consisting of a dashboard that provides the user with information and a service orientated architecture behind which provides the correspondent services to collect the required information. The development of such a business application and the background service infrastructure requires not only lots of know-how, but is also very expensive. This paper presents the MO-BIZZ platform that allows businesses to rapidly develop and offer new services to their customers. In fact, the MO-BIZZ platform enables autonomous organizations to deploy additional services and the combination and validation of existing cloud-based services from third-party provider. Thus, companies are able to save money for developing common business functionality and a complex background infrastructure for its provisioning. The challenge of this innovative technology is not only related to the development of a business model that shall facilitate rapid service conceptualization, but rather the service level agreements that need to be dynamically negotiated and accepted on different levels of the MO-BIZZ stack.