No abstract

No abstract

In cloud environments, resources are being shared between different organizations, and service providers of cloud resources require an efficient mechanism for metering resource usage to be able to charge customers fairly, based on actual resource utilization rather than resource rental time. One of the main challenges is the incompatibility of different usage data formats generated from different software and cloud resources, which need to be correlated and processed seamlessly to achieve a streamline metering process. In this paper, an architecture is proposed for collecting usage data and presenting it in a portable inter-operable format. Usage data is presented in Predictive Model Markup Language PMML, where data models exchange is adopted rather than the traditional data unification approaches. An extensible interpreted Cloud Metering Markup Language, CMML, is proposed for distributed usage data collection and processing. Finally, a prototype for CPU usage correlation is presented to demonstrate the applicability of the framework.

Big Data applications have demanding expectations on computational resources front. Thus, general purpose operating systems are not a good fit. In this paper, we present a new special purpose distributed micro-kernel designed with big data applications' needs in mind. The new micro-kernel adopts a core-based Asymmetric Multiprocessing (AMP) approach. It optimizes interrupt management and I/O to suit the Map-Reduce model. The proposed micro-kernel design is based on Inter-processor Interrupts over Ethernet (IPIoE) frames and a BareMetal Operating System Markup Language (BOSML). A transparent deployment mechanism is presented to completely shield the developer of the micro-kernel service from the underlying distribution infrastructure and decouple the application implementation from its deployment perspective. Based on the initial prototype and the experiments presented, a considerable gain in performance of average 2.34 folds was achieved using the distributed TeraSort benchmark over Linux/Hadoop.

Database tampering is a key security threat that impacts the integrity of sensitive in- formation of crucial businesses. The evolving risks of security threats as well as regulatory compliance are important driving forces for achieving better integrity and detecting pos- sible data tampering by either internal or external malicious perpetrators. We present DBKnot, an architecture for a tamper detection solution that caters to such problem while maintaining seamlessness and ease of retrofitting into existing append-only database ap- plications with near-zero modifications. We also pay attention to data confidentiality by making sure that the data never leaves the organization’s premises. We leverage designs like chains of record hashes to achieve the target solution. A set of preliminary exper- iments have been conducted that resulted in DBKnot adding an overhead equal to the original transaction time. We have run the same experimemts experiments with different parallelization and pipelining versions of DBKnot which resulted in cutting approximately 66% of the added overhead.